logo

WANDISCO FUSION® USER GUIDE

1. Introduction

Welcome to the User Guide for WANdisco Fusion, version 2.14.1.

1.1. What is Fusion?

WANdisco Fusion is a software application that allows Hadoop deployments to replicate HDFS data between Hadoop clusters that are running different, even incompatible versions of Hadoop. It is even possible to replicate between different vendor distributions and versions of Hadoop.

1.1.1. Benefits

  • Virtual File System for Hadoop, compatible with all Hadoop applications.

  • Single, virtual Namespace that integrates storage from different types of Hadoop, including CDH, HDP, EMC Isilon, Amazon S3/EMRFS and MapR.

  • Storage can be globally distributed.

  • WAN replication using the WANdisco Fusion LiveData platform, delivering single-copy consistent HDFS data, replicated between far-flung data centers.

1.2. Using this guide

This guide describes how to install and administer WANdisco Fusion as part of a multi data center Hadoop deployment, using either on-premises or cloud-based clusters. This guide contains the following:

Welcome

This chapter introduces this user guide and provides help with how to use it.

Release Notes

Details the latest software release, covering new features, fixes and known issues to be aware of.

Concepts

Explains the core concepts of how WANdisco Fusion operates, and how it fits into a Big Data environment.

Installation (On-premises)

Covers the steps required to install and set up WANdisco Fusion into a On-premises environment.

Installation (Cloud)

Covers the steps required to install and set up WANdisco Fusion into a Cloud-based environment.

Operation

The steps required to run, reconfigure and troubleshoot WANdisco Fusion.

Reference

Additional WANdisco Fusion documentation, including documentation for the available REST API.

1.3. Symbols in the documentation

In the guide we highlight types of information using the following call outs:

The alert symbol highlights important information.
The STOP symbol cautions you against doing something.
Tips are principles or practices that you’ll benefit from knowing or using.
The KB symbol shows where you can find more information, such as in our online Knowledge base.

1.4. Get support

See our online Knowledge base which contains updates and more information.

If you need more help raise a case on our support website.

We use terms that relate to the Hadoop ecosystem, WANdisco Fusion and WANdisco’s DConE replication technology. If you encounter any unfamiliar terms checkout the Glossary.

1.5. Local Language Support

WANdisco Fusion supports internationalization (i18n) and currently renders in the following languages.

Language

code

U.S. English

en-US

Simplified Chinese

zh-CN

During the command-line installation phase, the display language is set by the system’s locale. In use, the display language is determined through the user’s browser settings. Where language support is not available for your locale, then U.S. English will be displayed.

To handle non-ASCII characters in file and folder names, the LC_ALL environment variable must be set to en_US.UTF-8. This can be edited in /etc/wandisco/fusion/ui/main.conf. You must make sure that the locale is correctly installed.

There are a few areas which are not automatically translated, for example Email templates, but these can be easily modified as described in the relevant sections.

1.6. Give feedback

If you find an error or if you think some information needs improving, raise a case on our support website or email docs@wandisco.com.

2. Release Notes

2.1. Version 2.14.1 Build 3588

19 September 2019

For the release notes and information on Known issues, please visit the Knowledge base - WANdisco Fusion 2.14.1 Build 3588 Release Notes.

2.2. Version 2.14.0 Build 2675

4 July 2019

For the release notes and information on Known issues, please visit the Knowledge base - WANdisco Fusion 2.14.0 Build 2675 Release Notes.

3. Concepts

3.1. Product concepts

This section outlines some of the WANdisco Fusion concepts. More detailed information can be found in the Background Reading section of the Reference Guide.

3.1.1. What is WANdisco Fusion

WANdisco Fusion shares data between two or more clusters. Shared data is replicated between clusters using DConE, WANdisco’s proprietary coordination engine. This isn’t a spin on mirroring data, every cluster can write into the shared data directories and the resulting changes are coordinated in real-time between clusters.

100% Reliability

LiveData uses a set of Paxos-based algorithms to continue to replicate even after brief networks outages, data changes will automatically catch up once connectivity between clusters is restored.

Below the coordination stream, actual data transfer is done as an asynchronous background process and doesn’t consume MapReduce resources.

Replication where and when you need

WANdisco Fusion supports Selective replication, where you control which data is replicated to particular clusters, based on your security or data management policies. Data can be replicated globally if data is available to every cluster or just one cluster.

Fusion example
Figure 1. Simplified Replication diagram
The Benefits of WANdisco Fusion
  • Ingest data to any cluster, sharing it quickly and reliably with other clusters. Removing fragile data transfer bottlenecks, and letting you process data at multiple places improving performance and getting you more utilization from backup clusters.

  • Support a bimodal or multimodal architecture to enable innovation without jeopardizing SLAs. Perform different stages of the processing pipeline on the best cluster. Need a dedicated high-memory cluster for in-memory analytics? Or want to take advantage of an elastic scale-out on a cheaper cloud environment? Got a legacy application that’s locked to a specific version of Hadoop? WANdisco Fusion has the connections to make it happen. And unlike batch data transfer tools, WANdisco Fusion provides fully consistent data that can be read and written from any site.

  • Put away the emergency pager. If you lose data on one cluster, or even an entire cluster, WANdisco Fusion has made sure that you have consistent copies of the data at other locations.

  • Set up security tiers to isolate sensitive data on secure clusters, or keep data local to its country of origin.

  • Perform risk-free migrations. Stand up a new cluster and seamlessly share data using WANdisco Fusion. Then migrate applications and users at your leisure, and retire the old cluster whenever you’re ready.

3.2. WANdisco Fusion architecture

3.2.1. Example Workflow

The following diagram presents a simplified workflow for WANdisco Fusion, which illustrates a basic use case and points to how WANdisco’s distributed coordination engine (DConE) is implemented to overcome the challenges of coordination.

Example Workflow
Figure 2. Coordination
  1. User makes a request to create or change a file on the cluster.

  2. WANdisco Fusion coordinates File Open to the external cluster.

  3. File is added to underlying storage.

  4. WANdisco Fusion coordinates at configurable write increments and File Close with other clusters.

  5. WANdisco Fusion server at remote cluster pulls data from IHC server on source cluster.

  6. WANdisco Fusion server at remote site writes data to its local cluster.

3.2.2. Zones

A Zone represents the file system used in a standalone Hadoop cluster. Multiple Zones could be from separate clusters in the same data center, or could be from distinct clusters operating in geographically-separate data centers that span the globe. WANdisco Fusion operates as a distributed collection of servers. While each WANdisco Fusion server always belongs to only one Zone, a Zone can have multiple WANdisco Fusion servers (for load balancing and high availability). When you install WANdisco Fusion, you should create a Zone for each cluster’s file system.

3.2.3. Authorization and Authentication

The Fusion user interface provides an LDAP/AD connection, allowing Fusion users to be managed through a suitable Authorization Authority, such as an LDAP, Active Directory or Cloudera Manager-based system. Users can have their access to Fusion fine-tuned using assigned roles. Each Fusion user can be assigned one or more roles through the organization’s authorization authority, see Roles and Permissions for more information. You can also read more in the reference guide, or the operation section for details on connecting to LDAP/Active Directory.

Authz and Authn
Figure 3. Fusion Authorization and Authentication model

3.3. NameNode Proxy architecture

Fusion 2.14.1 introduced a new feature/deployment model called NameNode Proxy.

The NameNode Proxy provides an alternative to the WANdisco Fusion Client to replicate HDFS client calls. NameNode Proxy removes the need for Fusion client libraries by having Fusion coordinate activities performed against the Hadoop NameNode. This allows HDFS client actions made to the HDFS NameNode to be proxied and passed to the Fusion Server for replication.

Unlike the Fusion client, the NameNode Proxy does not require a client package install for every HDFS client node nor additional classpath changes for various HDFS supported applications. It simplifies the integration of WANdisco Fusion with Hadoop environments.

3.3.1. Example Workflow

NameNode Proxy Architecture
Figure 4. NameNode Proxy Architecture
A remote procedure call (RPC call) is a network protocol that is used for point-to-point communications between software applications. Client and server applications communicate during this process.
  1. The client sends an RPC call to the NameNode Proxy defined by the fs.defaultFS.

  2. NameNode Proxy will take the RPC call, create a proxy request and send it to the Fusion server.

  3. The local Fusion server sends the request to remote Fusion servers and waits for an agreed response.

  4. The NameNode Proxy will coordinate the request.

  5. Once the request is coordinated, the request is sent to the HDFS NameNode.

  6. The HDFS NameNode completes the request and sends a response back to the NameNode Proxy.

  7. The Fusion Server picks up the response from the NameNode Proxy Plugin.

  8. The Fusion Server sends the response to the NameNode Proxy as a NameNode Proxy response.

  9. The NameNode Proxy sends the protobuf response to the client.

3.3.2. Limitations

  • The NameNode Proxy node must be installed on a managed node with a HDFS gateway (a node with managed core-site.xml and hdfs-site.xml).

  • fs.fusion.push.threshold is not supported. Early pull requests will be sent out every time a block is written.

  • NameNode Proxy is a server-side process and so won’t recognise overrides from the Hadoop CLI client side configurations using -D.

  • NameNode Proxy does not support fsck.
    The fsck command finds the active NameNode and uses the http address associated with it to retrieve results. When using NameNode Proxy, the active NameNode address does not necessarily correlate with the real active NameNode and so the HTTP address of a standby NameNode can incorrectly be used instead. When running fsck, you should therefore point the command against the underlying NameNodes' namespace.

    hadoop fs -fsck hdfs://<underlyingNs>/

3.3.3. Supported Functionality

  • NameNode Proxy will operate with HDFS client commands. There is no impact on functional HDFS client behavior, however performance maybe effected.

  • Replication is supported for remote environments that use a non-HDFS file system.

  • NameNode Proxy supports Router Based Federation.

  • High Availability is supported for both HDFS NameNode High Availability (see Configure for High Availability Hadoop) and NameNode Proxy HA.

  • Fusion bypass is supported with NameNode Proxy.

  • NameNode Proxy supports operation with Live Hive with the appropriate configurations.

  • Kerberos and SSL security is supported.

3.3.4. Use Case for NameNode Proxy

The following use case illustrates how NameNode Proxy can be integrated into different types of deployments.

Single NameNode Proxy per zone
Figure 5. Single NameNode Proxy per zone

3.4. Deployment models

The following deployment models illustrate some of the common use cases for running WANdisco Fusion.

3.4.1. Analytic off-loading

In a typical on-premises Hadoop cluster, data ingest, analytic jobs all run through the same infrastructure where some activities impose a load on the cluster that can impact other activities. WANdisco Fusion allows you to divide up the workflow across separate environments, which lets you isolate the overheads associated with some events. You can ingest in one environment while using a different environment where capacity is provided to run the analytic jobs. You get more control over each environment’s performance.

  • You can ingest data from anywhere and query that at scale within the environment.

  • You can ingest data on premises (or where ever the data is generated) and query it at scale in another optimized environment, such as a cloud environment with elastic scaling that can be spun up only when queries jobs are queued. In this model, you may ingest data continuously but you don’t need to run a large cluster 24-hours-per-day for queries jobs.

3.4.2. Multi-stage jobs across multiple environments

A typical Hadoop workflow might involve a series of activities, ingesting data, cleaning data and then analyzing the data in a short series of steps. You may be generating intermediate output to be run against end-stage reporting jobs that perform analytical work, running all these work streams on a single cluster could require a lot of careful coordination with different types of workloads, conducting multi-stage jobs. This is a common chain of query activities for Hadoop applications, where you might ingest raw data, refine and augment it with other information, then eventually run analytic jobs against your output on a periodic basis, for reporting purposes, or in real-time.

In a replicated environment, however, you can control where those job stages are run. You can split this activity across multiple clusters to ensure the queries jobs needed for reporting purposes will have access to the capacity necessary to ensure that they run under within SLAs. You also can run different types of clusters to make more efficient use of the overall chain of work that occurs in a multi-stage job environments. You could have a cluster running that is tweaked and tuned for most efficient ingest, while running a completely different kind of environment that is tuned for another task, such as the end-stage reporting jobs that run against processed and augmented data. Running with Live data across multiple environments allows you to run each different type of activity in the most efficient way.

3.4.3. Migration

WANdisco Fusion allows you to move both the Hive data, stored in HCFS and associated Hive metadata from an on-premises cluster over to cloud-based infrastructure. There’s no need to stop your cluster activity; the migration can happen without impact to your Hadoop operations.

3.4.4. Disaster Recovery

As data is replicated between nodes on a continuous basis, WANdisco Fusion is an ideal solution for protecting your data from loss. If a disaster occurs, there’s no complicated switchover as the data is always operational.

3.4.5. Hadoop to S3

WANdisco Fusion can be used to migrate or replicate data from a Hadoop platform to S3, or S3 compatible, storage. WANdisco’s S3 plugin provides:

  • LiveData transactional replication from the on-premise cluster to an S3 bucket

  • Consistency check of data between the Hadoop platform and the S3 bucket

  • Point-in-time batch operations to return to consistency from Hadoop to S3

  • Point-in-time batch operations to return to consistency from S3 back to Hadoop

However it does not provide any facility for LiveData transactional replication from S3 to Hadoop.

3.5. Working in the Hadoop ecosystem

This section covers the final step in setting up a WANdisco Fusion cluster, where supported Hadoop applications are plugged into WANdisco Fusion’s synchronized distributed namespace. It won’t be possible to cover all the requirements for all the third-party software covered here, we strongly recommend that you get hold of the corresponding documentation for each Hadoop application before you work through these procedures.

3.5.1. Application Plugins

Multiple plugins allow you to extend the WANdisco Fusion platform to support the live replication of Hadoop applications. See docs.wandisco.com for the plugin user guides.

Fusion Plugin for Live Hive

This enables WANdisco Fusion to replicate Apache Hive’s metastore, allowing WANdisco Fusion to maintain a replicated instance of Hive’s metadata.

Fusion Plugin for Live Ranger

Apache Ranger is a centralized security console for Hadoop clusters, a preferred solution for Hortonworks HDP. Use the Fusion Plugin for Live Ranger to extend the capabilities of WANdisco Fusion to Apache Ranger across multiple Hadoop environments, and keep your security policies consistent.

Fusion Plugin for Live Sentry

Apache Sentry provides fine-grained authorization and role-based access control on the Cloudera platform. Use the Fusion Plugin for Live Sentry to extend WANdisco Fusion with the ability to replicate policies among Apache Sentry Policy Provider instances.

3.5.2. Hadoop File System Configuration

The following section explains how Fusion interacts with and replicates to file systems and object stores.

There are several options available for configuring Hadoop clients to work with Fusion, with different configurations suiting different types of deployment. Configuration is done during installation but can also be changed in-life through the Fusion UI, or even manually by amending the core-site for the environment. Manual amendments to the core-site should only be required for some unmanaged environments (such as S3, LocalFs, etc).

Fusion Client
  1. The Hadoop file system looks at either the input URI (e.g. fusion://) or the fs.defaultFS property for a scheme.

  2. It then looks for the implementation property, for example, fs.fusion.impl to instantiate a filesystem.

  3. In order to instantiate Fusion, the implementation property needs to match with a compatible implementation, for example, Fusion’s HDFS class fs.fusion.impl=com.wandisco.fs.client.FusionHdfs.

  4. Fusion uses the fs.underlyingClass property to identify the underlying filesystem, therefore, it can map local client requests to the underlying filesystem.

  5. The consequence of using this URI or fs.defaultFS scheme is that Fusion can intercept client requests to replicate to another filesystem, as well as processing requests for the underlying filesystem.

NameNode Proxy with HDFS
  • As of Fusion 2.14.1, the NameNode Proxy is available as an alternative to the Fusion Client on Hadoop clusters utilising HDFS.

  • The fs.underlyingClass property is still used to identify the underlying filesystem, however, the implementation property (e.g fs.fusion.impl or fs.hdfs.impl) is no longer required.

  • The NameNode Proxy intercepts client and application requests to the NameNode(s) directly. See NameNode Proxy Architecture for more detailed information.

Universal Resource Identifier (URI)

Client access to Fusion is chiefly driven by URI selection. The exception to this being the NameNode Proxy, which is explained in the further detail in the Fusion’s URI options section.

The Hadoop URI consists of a scheme, authority, and path. The scheme and authority together determines the FileSystem implementation.

For example, the default for on-premise Hadoop clusters is HDFS and is referred to by the fs.hdfs.impl property. This property points at the Java class that handles references to files under the hdfs:// prefix.

This prefix is entirely arbitrary, and you could use any prefix that you want, providing that it points to an appropriate fs.*.impl that will handle the filesystem commands that you need.

MapR must use WANdisco’s native fusion:// URI, instead of the default hdfs://. See Fusion URI with HCFS option in the section below.
On-premises - Fusion options for URI selection

Namenode Proxy with HDFS
This option was introduced in Fusion 2.14.1. The NameNode Proxy with HDFS option provides an alternative to the Fusion client to replicate HDFS client calls.

Fusion NameNode Proxy removes the need for Fusion client libraries by having Fusion coordinate activities performed against the Hadoop NameNode. This allows HDFS client actions made to the HDFS NameNode to be proxied and passed to the Fusion Server for replication.

Unlike the other URI Selection options (which all use the Fusion client), the NameNode Proxy does not require a client package install for every HDFS client node nor additional classpath changes for various HDFS supported applications. It simplifies the integration of WANdisco Fusion with Hadoop environments.

HDFS URI with HDFS
This option allows WANdisco Fusion to operate entirely within the HDFS standard URI scheme (hdfs://). As such, application and client requests do not need to reference a different URI scheme in order to replicate files and directories.

This can also be useful if the Hadoop applications do not support a custom URI scheme, such as the WANdisco Fusion URI, or HCFS standards.

This configuration will not allow paths with the fusion:// URI to be used; only paths starting with hdfs:// that correspond to a mapped path will be replicated. The underlying file system will be an instance of the HDFS DistributedFileSystem, which will support applications that are not written to the HCFS specification.

Fusion URI with HCFS
When this option is selected, you need to use fusion:// URI for all data that must be replicated over an instance of the Hadoop Compatible File System. If your deployment includes Hadoop applications that are either unable to support the Fusion URI or are not written to the HCFS specification, this option will not work.

Fusion URI with HDFS
This differs from the HCFS variant in that while the WANdisco Fusion URI is used to identify data to be replicated, the replication is performed using HDFS itself. This option can be used if you are deploying applications that can support the WANdisco Fusion URI but not the Hadoop Compatible File System.

Fusion URI and HDFS URI with HDFS
This "mixed mode" supports all the replication schemes (fusion://, hdfs:// and no scheme) and uses HDFS for the underlying file system, to support applications that aren’t written to the HCFS specification.

Cloud - Fusion options for URI selection

Fusion URI with HCFS
When this option is selected, you need to use fusion:// URI for all data that must be replicated over an instance of the Hadoop Compatible File System.

Platforms that must run with Fusion URI with HCFS

  • Azure

  • LocalFS

  • UmanagedBigInsights

  • UnmanagedSwift

  • UnmanagedGoogle

  • UnmanagedS3

  • UnmanagedEMR

  • MapR

Default filesystem
This option requires manual configuration of the environment’s filesystem implementation so that it references Fusion’s HCFS class. Much like the HDFS URI with HDFS scheme for on-premise options, this will integrate Fusion into the default URI scheme for the environment.

Examples
  1. On ADL, Fusion would require fs.adl.impl to reference the Fusion HCFS class, instead of the default ADL class.

  2. On WASB, Fusion would require fs.wasb.impl to reference the Fusion HCFS class, instead of the default WASB class.

Implementation property

The fs.*.impl property is the abstract FileSystem implementation that will be used.

<property>
  <name>fs.<implementation-name>.impl</name>
<value>valid.impl.class.for.an.hcfs.implementation</value>
</property>
Further explanation of properties

If you would like to understand more about the properties referenced in this section, please refer to the background reading section for this segment.

3.5.3. Hive

This section provides guidance on integrating WANdisco Fusion with Apache Hive when using the specific fusion:/// URI scheme.

It aims to accomplish the following goals:

  • Replicate Hive table storage.

  • Use fusion URIs as store paths.

  • Use fusion URIs as load paths.

Prerequisites:

  • Knowledge of Hive architecture.

  • Ability to modify Hadoop site configuration.

  • WANdisco Fusion installed and operating.

Replicating Hive Storage via fusion:///

The following requirements come into play if you have deployed WANdisco Fusion using with its native fusion:/// URI.

In order to store a Hive table, you specify a Fusion URI when creating a table. E.g. consider creating a table called log that will be stored in a replicated directory.

CREATE TABLE log(requestline string) stored as textfile location 'fusion:///repl1/hive/log';. Note: Replicating table storage without sharing the Hive metadata will create a logical discrepancy in the Hive catalog. For example, consider a case where a table is defined on one cluster and replicated on the HCFS to another cluster. A Hive user on the other cluster would need to define the table locally in order to make use of it.

Don’t use namespace
Make sure you don’t use the namespace name e.g. use fusion:///user/hive/log, not fusion://nameserviceA/user/hive/log.
Replicated directories as store paths

It’s possible to configure Hive to use WANdisco Fusion URIs as output paths for storing data, to do this you must specify a Fusion URI when writing data back to the underlying Hadoop-compatible file system (HCFS). For example, consider writing data out from a table called log to a file stored in a replicated directory:

INSERT OVERWRITE DIRECTORY 'fusion:///repl1/hive-out.csv' SELECT * FROM log;
Replicated directories as load paths

In this section we’ll describe how to configure Hive to use fusion URIs as input paths for loading data.

It is not common to load data into a Hive table from a file using the fusion URI. When loading data into Hive from files the core-site.xml setting fs.defaultFS must also be set to fusion, which may not be desirable. It is much more common to load data from a local file using the LOCAL keyword:

LOAD DATA LOCAL INPATH '/tmp/log.csv' INTO TABLE log;

If you do wish to use a fusion URI as a load path, you must change the fs.defaultFS setting to use WANdisco Fusion, as noted in a previous section. Then you may run:

LOAD DATA INPATH 'fusion:///repl1/log.csv' INTO TABLE log;

3.5.4. Impala

Prerequisites
  • Knowledge of Impala architecture.

  • Ability to modify Hadoop site configuration.

  • WANdisco Fusion installed and operating.

Impala Parcel

If you plan to use WANdisco Fusion’s own fusion:/// URI, then you will need to use the provided parcel (see the screenshot, below for link in the Client Download section of the Settings screen):

WANdisco Fusion tree
Figure 6. Impala Parcel link

Follow the same steps described for installing the WANdisco Fusion client, downloading the parcel and SHA file, i.e.:

  1. Have cluster with CDH installed with parcels and Impala.

  2. Copy the FUSION_IMPALA parcel and SHA into the local parcels repository, on the same node where Cloudera Manager Services is installed, this need not be the same location where the Cloudera Manager Server is installed. The default location is at: /opt/cloudera/parcel-repo, but is configurable. In Cloudera Manager, you can go to the Parcels Management Page → Edit Settings to find the Local Parcel Repository Path. See Parcel Locations.

    FUSION_IMPALA should be available to distribute and activate on the Parcels Management Page, remember to click Check for New Parcels button.

  3. Once installed, restart the cluster.

  4. Impala reads on Fusion files should now be available.

Setting the CLASSPATH

In order for Impala to load the Fusion Client jars, the user needs to make a small configuration change in their Impala service, through Cloudera Manager. In Cloudera Manager, the user needs to add an environment variable in the section Impala Service Environment Advanced Configuration Snippet (Safety Valve).

AUX_CLASSPATH='colon-delimited list of all the Fusion client jars'

The following command gives an example of how to do this.

echo "AUX_CLASSPATH=$((for i in /opt/cloudera/parcels/FUSION/lib/*.jar; do echo -n "${i}:"; done) | sed 's/\:$//g')"
WANdisco Fusion tree
Figure 7. Classpath configuration - Impala

3.5.5. Presto

Presto Interoperability

Presto is an open source distributed SQL query engine for running interactive analytic queries. It can query and interact with multiple data sources, and can be extended with plugins.

Presto requires the use of Java 8 and has internal dependencies on Java library versions that may conflict with those of the Hadoop distribution with which it communicates when using the “hive-hadoop2” plugin.

Presto and Fusion

WANdisco Fusion leverages a replacement client library when overriding the hdfs:// scheme for access to the cluster file system in order to coordinate that access among multiple clusters. This replacement library is provided in a collection of jar files in the /opt/wandisco/fusion/client/lib directory for a standard installation. These jar files need to be available to any process that accesses the file system using the com.wandisco.fs.client.FusionHdfs implementation of the Apache Hadoop FileSystem API.

Because Presto requires these classes to be available to the hive-hadoop2 plugin, they must reside in the plugin/hive-hadoop2 directory of the Presto installation.

Using the Fusion Client Library with Presto
  1. Copy the JAR files in the plugin/hive-hadoop2 directory of each Presto server.

  2. Restart the Presto coordinators.

It is also important to confirm that the Presto configuration includes the necessary properties to function correctly with the hive-hadoop2 plugin.

The specific values below will need to be adjusted for the actual environment, including references to the WANdisco replicated metastore, the HDP cluster configuration that includes Fusion configuration, and Kerberos-specific information to allow Presto to interoperate with a secured cluster.

connector.name=hive-hadoop2
hive.metastore.uri=thrift://presto02-vm1.test.server.com:9084
hive.config.resources=/etc/hadoop/conf/core-site.xml,/etc/hadoop/conf/hdfs-site.xml
hive.metastore.authentication.type=KERBEROS
hive.metastore.service.principal=hive/presto02-vm1.test.server.com@WANDISCO.HADOOP
hive.metastore.client.principal=presto/presto02-vm0.test.server.com@WANDISCO.HADOOP
hive.metastore.client.keytab=/etc/security/keytabs/presto.keytab
hive.hdfs.authentication.type=KERBEROS
hive.hdfs.impersonation.enabled=true
hive.hdfs.presto.principal=hdfs-presto2@WANDISCO.HADOOP
hive.hdfs.presto.keytab=/etc/security/keytabs/hdfs.headless.keytab

Keytabs and principals will need to be configured correctly, and as the hive-hadoop2 Presto plugin uses YARN for operation, the /user/yarn directory must exist and be writable by the yarn user in all clusters in which Fusion operates.

Known Issue

Presto embeds Hadoop configuration defaults into the hive-hadoop2 plugin, including a core-default.xml file that specifies the following property entry:

<property>
  <name>hadoop.security.authentication</name>
  <value>simple</value>
  <description>Possible values are simple (no authentication), and kerberos
  </description>
</property>

Although Presto allows the hive-hadoop2 plugin to use additional configuration properties by adding entries like the following in a .properties file in the etc/catalog directory:

hive.config.resources=/etc/hadoop/conf/core-site.xml,/etc/hadoop/conf/hdfs-site.xml

This entry allows extra configuration properties to be loaded from a standard Hadoop configuration file, but those entries cannot override settings that are embedded in the core-default.xml that ships with the Presto hive-hadoop2 plugin.

In a kerberized implementation the Fusion client library relies on the ability to read the hadoop.security.authentication configuration property to determine if it should perform a secure handshake with the Fusion server. Without that property defined, the client and server will fail to perform their security handshake, and Presto queries will not succeed.

Workaround

The solution to this issue is to update the core-default.xml file contained in the hive-hadoop2 plugin:

$ mkdir ~/tmp
$ cd ~/tmp
$ jar -xvf <path to…>/presto-server-0.164/plugin/hive-hadoop2/hadoop-apache2-0.10.jar

Edit the core-default.xml file to update the hadoop.security.authentication property so that its value is “kerberos”

$ Jar -uf <path to...>/presto-server-0.164/plugin/hive-hadoop2/hadoop-apache2-0.10.jar core-default.xml

Distribute the hadoop-apache2-0.10.jar to all Presto nodes, and restart the Presto coordinator.

3.5.6. Oozie

NameNode Proxy integration

These steps are required on a Hadoop cluster running Oozie and NameNode Proxy (i.e. not required for Fusion Client installations).

Known issue - additional configuration for compatibility with NameNode Proxy

If the NameNodes are not listed as HDFS servers when using Oozie and NameNode Proxy, Oozie will never attempt to create a delegation token with a NameNode.

As such, some additional configuration is required for NameNode Proxy to work with Oozie (please note the examples below assume that the NameNode Proxy is referenced as a nameservice):

The examples below refer to the NameNode Proxy as a nameservice, which is the case when you have deployed more than one NameNode Proxy and they are configured for High Availability. If there is only one NameNode Proxy present on the cluster, adjust the nameservice example to the NameNode Proxy hostname:port.
  1. Add Namenode proxy to the Oozie whitelist in oozie-site.xml:

    Ambari = Custom oozie-site
    Cloudera = Oozie Server Advanced Configuration Snippet (Safety Valve) for oozie-site.xml

    oozie.service.HadoopAccessorService.nameNode.whitelist = <NameNode_Proxy-nameservice>,<NameNode-nameservice>
  2. Make NameNode Proxy the default NameNode in oozie-site.xml:

    Ambari = Custom oozie-site
    Cloudera = Oozie Server Advanced Configuration Snippet (Safety Valve) for oozie-site.xml

    oozie.actions.default.name-node = <NameNode_Proxy-nameservice>
  3. In any configured workflow.xml on HDFS, ensure that both NameNode Proxy and NameNode nameservices are recognized as HDFS services:

    mapreduce.job.hdfs-servers = <NameNode_Proxy-nameservice>,<NameNode-nameservice>

See Apache Oozie and Oozie Workflow for official documentation of these properties.

Ambari / Hortonworks (HDP) Fusion Client integration

In Ambari / Hortonworks (HDP) you need to apply the following procedure if Oozie was installed after completing the WANdisco Fusion installation.

Not required if using the NameNode Proxy
If the Hadoop cluster is using the NameNode Proxy rather than the Fusion Client, then these symlinks are not required.
  1. Open a terminal to the node with root privileges.

  2. If Fusion was previously installed and has now been removed, check that any dead symlinks have been removed.

    cd /usr/hdp/current/oozie-server/libext
    ls -l
    rm [broken symlinks]
  3. Create the symlinks for fusion client jars.

    ln -s /opt/wandisco/fusion/client/jars/* /usr/hdp/current/oozie-server/libext
  4. In Ambari, stop the Oozie Server service.

  5. Open a terminal session as user oozie and run:

    /usr/hdp/current/oozie-server/bin/oozie-setup.sh prepare-war
  6. In Ambari, start the Oozie Server service.

    It is worth noting that the new symlinks get created, but if previous symlinks have not been manually removed first, the war packaging (which happens when oozie server is started) will fail. This will cause the Oozie Server startup to fail.

    You need to ensure old symlinks in /usr/hdp/current/oozie-server/libext are removed before installing a new client stack.

Manual Client installations (instead of Stacks)

Installing clients via RPM/Deb packages will not stop and repackage Oozie. If Oozie was running prior to the client installation, you will need to manually stop Oozie, then Oozie setup command:

oozie-setup.sh prepare-war

If possible, complete these actions through Ambari.

In Ambari, using the "Refresh configs" service action on the WANdisco Fusion service should trigger re-linking and the prepare-war process.

If not installed directly via RPM/Deb packages, you should use the manual process for reinstalling the package, followed by the same steps noted above to stop and restart Oozie, using the setup script.

3.5.7. Oracle: Big Data Appliance

Each node in an Oracle:BDA deployment has multiple network interfaces, with at least one used for intra-rack communications and one used for external communications. WANdisco Fusion requires external communications so configuration using the public IP address is required instead of using host names.

Prerequisites
  • Knowledge of Oracle:BDA architecture and configuration.

  • Ability to modify Hadoop site configuration.

Required steps
  • Configure WANdisco Fusion to support Kerberos. See Setting up Kerberos

  • Configure WANdisco Fusion to work with NameNode High Availability described in Oracle’s documentation

  • Restart the cluster, WANdisco Fusion and IHC processes. See init.d management script

  • Test that replication between zones is working.

Operating in a multi-homed environment

Oracle:BDA is built on top of Cloudera’s Hadoop and requires some extra steps to support multi-homed network environment.

3.5.8. Apache Livy

There’s an issue with running Apache Livy. As a Spark1 application, it does not use the standard Hadoop classpath, but also does not use the Spark Assembly. Livy may fail to start with FusionHdfs class not found.

Based on the current active version of HDP/Livy, you can resolve this with the following symlink.

 ln -s /opt/wandisco/fusion/client/lib/* /usr/hdp/current/livy-server/jars/
Not required if using the NameNode Proxy
If the Hadoop cluster is using the NameNode Proxy rather than the Fusion Client, then this symlink is not required.

3.5.9. Apache Tez

Apache Tez is a YARN application framework that supports high performance data processing through DAGs. When set up, Tez uses its own tez.tar.gz containing the dependencies and libraries that it needs to run DAGs.

Tez with Hive

In order to make Hive with Tez work, you need to append the Fusion jar files in tez.cluster.additional.classpath.prefix under the Advanced tez-site section:

tez.cluster.additional.classpath.prefix = /opt/wandisco/fusion/client/lib/*

e.g. WANdisco Fusion tree

Running Hortonworks Data Platform, the tez.lib.uris parameter defaults to /hdp/apps/${hdp.version}/tez/tez.tar.gz. So, to add Fusion libs, there are three options.

Fusion installer users Option 1

Option 1: Delete the tez.lib.uris path, e.g. "/hdp/apps/${hdp.version}/tez/tez.tar.gz". Instead, use a list including the path where the tez.tar.gz file will unpack, and the path where Fusion libs are located.

WANdisco Fusion Text
Figure 8. Specify additional classpath info for Fusion libs

Option 2: Or unpack tez.tar.gz, repack with WANdisco Fusion libs and re-upload to HDFS.

Option 3: Alternatively, you may set the tez.lib.uris property with the path to the WANdisco Fusion client jar files, e.g.

<property>
  <name>tez.lib.uris</name>
# Location of the Tez jars and their dependencies.
# Tez applications download required jar files from this location, so it should be public accessible.
  <value>${fs.default.name}/apps/tez/,${fs.default.name}/apps/tez/lib/</value>
</property>
All these methods are vulnerable to a platform (HDP) upgrade.

3.5.10. Tez / Hive2 with LLAP

The following configuration changes are needed when running Tez with Low Latency Analytical Processing functionality.

Tez Overview

You can read about the results of testing Hive2 with LLAP - Low Latency Analytical Processing, using Apache Slider to run Tez Application Masters on YARN. Inevitably, running a Tez query through this interface results in a FusionHDFS class not found.

The following steps show an example remedy, through the bundling of the client jars into the tez.lib.uris tar.gz.

Verified on HDP 2.6.2
The following example is tested on HDP 2.6.2. The procedure may alter on different platforms.
  1. First, extract existing Tez library to a local directory.

    # mkdir /tmp/tezdir
    # cd /tmp/tezdir
    # cp /usr/hdp/2.6*/tez_hive2/lib/tez.tar.gz .
    # tar xvzf tez.tar.gz
  2. Add the Fusion client jars to the same extracted location.

    # cp /opt/wandisco/fusion/client/lib/* .
  3. Re-package the Tez library including the Fusion jars.

    # tar cvzf tez.tar.gz *
  4. Upload the enlarged Tez library to HDFS (taking a backup of original).

    # hdfs dfs -cp /hdp/apps/<your-hdp-version>/tez_hive2/tez.tar.gz /user/<username>/tez.tar.gz.pre-WANdisco
    # hdfs dfs -put tez.tar.gz /hdp/apps/<your-hdp-version>/tez_hive2/

    Note The <your-hdp-version> component of the path name needs to match the point release of HDP you are using. This should be in the form 2.major.minor.release-build id e.g. /hdp/apps/2.6.3.0-235/tez_hive2

  5. Restart LLAP service through Ambari.

3.5.11. Apache Kafka

Apache Kafka is a distributed publish-subscribe messaging system. Now part of the Apache project, Kafka is fast, scalable and by its nature, distributed, either across multiple servers, clusters or even data centers. See Apache Kafka.

Known problem
When Ranger auditing is enabled for Kafka, the audit logging data spools on local disk because the write to HDFS fails. The failure is caused by a "no class found" issue with the Fusion client. A typical error message if you added the Fusion client jars location to the CLASSPATH:
java.lang.ClassCastException: com.wandisco.fs.client.FusionHdfs cannot be cast to org.apache.hadoop.fs.FileSystem error.

workaround
In order to override the fs.hdfs.impl configuration in core-site.xml, all that we need to do is to add a custom property in Custom ranger-kafka-audit under Kafka Config in Ambari.

  1. Ambari → Kafka → Configs

  2. Expand Custom ranger-kafka-audit

  3. Add the following property:

    xasecure.audit.destination.hdfs.config.fs.hdfs.impl=org.apache.hadoop.hdfs.DistributedFileSystem
  4. Save the changes.

3.5.12. Solr

Apache Solr is a scalable search engine that can be used with HDFS. In this section we cover what you need to do for Solr to work with a WANdisco Fusion deployment.
Note: Solr only comes with CDH and IOP 4.2 and greater.

Minimal deployment using the default hdfs:// URI

Getting set up with the default URI is simple, Solr just needs to be able to find the fusion client jar files that contain the FusionHdfs class.

  1. Copy the Fusion/Netty jars into the classpath. Please follow these steps on all deployed Solr servers. For CDH5.4 with parcels, use these two commands:

    cp /opt/cloudera/parcels/FUSION/lib/fusion* /opt/cloudera/parcels/CDH/lib/solr/webapps/solr/WEB-INF/lib
    cp /opt/cloudera/parcels/FUSION/lib/netty-all-*.Final.jar /opt/cloudera/parcels/CDH/lib/solr/webapps/solr/WEB-INF/lib
    cp /opt/cloudera/parcels/FUSION/lib/wd-guava-15.0.jar /opt/cloudera/parcels/CDH/lib/solr/webapps/solr/WEB-INF/lib
    cp /opt/cloudera/parcels/FUSION/lib/bcprov-jdk15on-1.54.jar /opt/cloudera/parcels/CDH/lib/solr/webapps/solr/WEB-INF/lib
  2. Restart all Solr Servers.

  3. Solr is now successfully configured to work with WANdisco Fusion.

Minimal deployment using the WANdisco "fusion://" URI

This is a minimal working solution with Solr on top of fusion.

Requirements

Solr will use a shared replicated directory.

Not required if using the NameNode Proxy
If the Hadoop cluster is using the NameNode Proxy rather than the Fusion Client, then this symlink is not required.
  1. Symlink the WANdisco Fusion jars into Solr webapp.

    cd /opt/cloudera/parcels/CDH/lib/solr/webapps/solr/WEB-INF/lib
    ln -s /opt/cloudera/parcels/FUSION/lib/fusion* .
    ln -s /opt/cloudera/parcels/FUSION/lib/netty-all-4* .
    ln -s /opt/cloudera/parcels/FUSION/lib/bcprov-jdk15on-1.52 .
  2. Restart Solr.

  3. Create instance configuration.

    $ solrctl instancedir --generate conf1
  4. Edit conf1/conf/solrconfig.xml and replace solr.hdfs.home in directoryFactory definition with actual fusion:/// uri, like fusion:///repl1/solr

  5. Create solr directory and set solr:solr permissions on it.

    $ sudo -u hdfs hdfs dfs -mkdir fusion:///repl1/solr
    $ sudo -u hdfs hdfs dfs -chown solr:solr fusion:///repl1/solr
  6. Upload configuration to zk.

    $ solrctl instancedir --create conf1 conf1
  7. Create collection on first cluster.

    $ solrctl collection --create col1 -c conf1 -s 3
Tip
For Cloudera, fs.hdfs.impl.disable.cache = true should be set for Solr servers. (don’t set this options cluster-wide, that will stall the WANdisco Fusion server with an unbounded number of client connections).

3.5.13. Flume

This set of instructions will set up Flume to ingest data via the fusion:///` URI.

Edit the configuration, set "agent.sources.flumeSource.command" to the path of the source data. Set “agent.sinks.flumeHDFS.hdfs.path” to the replicated directory of one of the DCs. Make sure it begins with fusion:/// to push the files to Fusion and not hdfs.

Prerequisites
  • Create a user in both the clusters 'useradd -G hadoop <username>'

  • Create user directory in hadoop fs 'hadoop fs -mkdir /user/<username>'

  • Create replication directory in both DC’s 'hadoop fs -mkdir /fus-repl'

  • Set permission to replication directory 'hadoop fs -chown username:hadoop /fus-repl'

  • Install and configure WANdisco Fusion.

Setting up Flume through Cloudera Manager

If you want to set up Flume through Cloudera Manager follow these steps:

  1. Download the client in the form of a parcel and the parcel.sha through the UI.

  2. Put the parcel and .sha into /opt/cloudera/parcel-repo on the Cloudera Managed node.

  3. Go to the UI on the Cloudera Manager node. On the main page, click the small button that looks like a gift wrapped. box and the FUSION parcel should appear (if it doesn’t, try clicking Check for new parcels and wait a moment).

  4. Install, distribute, and activate the parcel.

  5. Repeat steps 1-4 for the second zone.

  6. Make sure replicated rules are created for sharing between Zones.

  7. Go onto Cloudera Manager’s UI on one of the zones and click Add Service.

  8. Select the Flume Service. Install the service on any of the nodes.

  9. Once installed, go to Flume→Configurations.

  10. Set System User to hdfs

  11. Set Agent Name to agent

  12. Set Configuration File to the contents of the flume.conf configuration.

  13. Restart Flume Service.

  14. Selected data should now be in Zone1 and replicated in Zone2

  15. To check data was replicated, open a terminal onto one of the DCs and become hdfs user, e.g. su hdfs, and run.

    hadoop fs -ls /repl1/flume_out"
  16. On both Zones, there should be the same FlumeData file with a long number. This file will contain the contents of the source(s) you chose in your configuration file.

3.5.14. Spark1

It’s possible to deploy WANdisco Fusion with Apache’s high-speed data processing engine. Note that prior to version 2.9.1 you needed to manually add the SPARK_CLASSPATH.

Spark with CDH

There is a known issue where Spark is not picking up hive-site.xml, So that Hadoop configuration is not localised when submitting job in yarn-cluster mode (Fixed in version Spark 1.4).

You need to manually add it in by either:

  • Copy /etc/hive/conf/hive-site.xml into /etc/spark/conf.
    or

  • Do one of the following, depending on which deployment mode you are running in:

    Client

    set HADOOP_CONF_DIR to /etc/hive/conf/ (or the directory where hive-site.xml is located).

    Cluster

    add --files=/etc/hive/conf/hive-site.xml (or the path for hive-site.xml) to the spark-submit script.

  • Deploy configs and restart services.

Using the FusionUri
The fusion:/// URI has a known issue where it complains about "Wrong fs". For now Spark is only verified with FusionHdfs going through the hdfs:/// URI.
Fusion Spark Interoperability

Spark applications are run on a cluster as independent sets of processes, coordinated by the SparkContext object in the driver program. To run on a cluster, the SparkContext can connect to several types of cluster managers (either Spark’s own standalone cluster manager, Mesos or YARN), which allocate resources across applications. Once connected, Spark acquires executors on nodes in the cluster, which are processes that run computations and store data for your application. Next, it sends your application code (defined by JAR or Python files passed to SparkContext) to the executors. Finally, SparkContext sends tasks to the executors to run.

Spark
Figure 9. Spark operation
Spark and Fusion

WANdisco Fusion uses a replacement client library when overriding the hdfs:// scheme for access to the cluster file system in order to coordinate that access among multiple clusters. This replacement library is provided in a collection of jar files in the /opt/wandisco/fusion/client/lib directory for a standard installation. These jar files need to be available to any process that accesses the file system using the com.wandisco.fs.client.FusionHdfs implementation of the Apache Hadoop File System API.

Because Spark does not provide a configurable mechanism for making the Fusion classes available to the Spark history server, the Spark Executor or Spark Driver programs, WANdisco Fusion client library classes need to be made available in the existing Spark assembly jar that holds the classes used by these Spark components. This requires updating that assembly jar to incorporate the Fusion client library classes.

Updating the Spark Assembly JAR

This is one of a number of methods that may be employed to provide Fusion-Spark integration. We hope to cover some alternate methods at a later date.

Hortonworks HDP
  1. First, make a backup of the original Spark assembly jar:

    $ cp /usr/hdp/<version>/spark/lib/spark-assembly-<version>-hadoop<version>.jar /usr/hdp/<version>/spark/lib/spark-assembly-<version>-hadoop<version>.jar.original

    Then follow this process to update the Spark assembly jar.

    $ mkdir /tmp/spark_assembly
    $ cd /tmp/spark_assembly
    $ jar -xf /opt/wandisco/fusion/client/lib/bcprov-jdk15on.jar
    $ jar -xf /opt/wandisco/fusion/client/lib/fusion-adk-common.jar
    $ jar -xf /opt/wandisco/fusion/client/lib/fusion-adk-netty.jar
    $ jar -xf /opt/wandisco/fusion/client/lib/fusion-client-common.jar
    $ jar -xf /opt/wandisco/fusion/client/lib/fusion-common.jar
    $ jar -xf /opt/wandisco/fusion/client/lib/wd-guava.jar
    $ jar -xf /opt/wandisco/fusion/client/lib/fusion-adk-client.jar
    $ jar -xf /opt/wandisco/fusion/client/lib/fusion-adk-hadoop.jar
    $ jar -xf /opt/wandisco/fusion/client/lib/fusion-adk-security.jar
    $ jar -xf /opt/wandisco/fusion/client/lib/fusion-client-hdfs.jar
    $ jar -xf /opt/wandisco/fusion/client/lib/fusion-messaging-core.jar
    $ jar -xf /opt/wandisco/fusion/client/lib/wd-netty-all.jar
    
    jar -uf /usr/hdp/<version>/spark/lib/spark-assembly-<version>-hadoop<version>.jar com/** org/** META-INF/**
  2. You now have both the original Spark assembly jar (with the extension “.original”) and a version with the Fusion client libraries available in it. The updated version needs to be made available on each node in the cluster in the /usr/hdp/<version>/spark/lib directory.

  3. If you need to revert to the original Spark assembly jar, simply copy it back in place on each node in the cluster.

Cloudera CDH

The procedure for Cloudera CDH is much the same as the one for HDP, provided above. Note that path differences:

  1. First, make a backup of the original Spark assembly jar:

    $ cp /opt/cloudera/parcels/CDH-<version>.cdh<version>/jars/spark-assembly-<version>-cdh<version>-hadoop<version>-cdh<version>.jar /opt/cloudera/parcels/CDH-<version>.cdh<version>/jars/spark-assembly-<version>-cdh<version>-hadoop<version>-cdh<version>.jar.original

    Then follow this process to update the Spark assembly jar.

    $ mkdir /tmp/spark_assembly
    $ cd /tmp/spark_assembly
    jar -xf /opt/cloudera/parcels/FUSION-<fusion-version>-cdh<version>/lib/bcprov-jdk15on-1.54.jar
    jar -xf /opt/cloudera/parcels/FUSION-<fusion-version>-cdh<version>/lib/fusion-adk-common.jar
    jar -xf /opt/cloudera/parcels/FUSION-<fusion-version>-cdh<version>/lib/fusion-adk-netty.jar
    jar -xf /opt/cloudera/parcels/FUSION-<fusion-version>-cdh<version>/lib/fusion-client-common.jar
    jar -xf /opt/cloudera/parcels/FUSION-<fusion-version>-cdh<version>/lib/fusion-common.jar
    jar -xf /opt/cloudera/parcels/FUSION-<fusion-version>-cdh<version>/lib/wd-guava.jar
    jar -xf /opt/cloudera/parcels/FUSION-<fusion-version>-cdh<version>/lib/fusion-adk-client.jar
    jar -xf /opt/cloudera/parcels/FUSION-<fusion-version>-cdh<version>/lib/fusion-adk-hadoop.jar
    jar -xf /opt/cloudera/parcels/FUSION-<fusion-version>-cdh<version>/lib/fusion-adk-security.jar
    jar -xf /opt/cloudera/parcels/FUSION-<fusion-version>-cdh<version>/lib/fusion-client-hdfs.jar
    jar -xf /opt/cloudera/parcels/FUSION-<fusion-version>-cdh<version>/lib/fusion-messaging-core.jar
    jar -xf /opt/cloudera/parcels/FUSION-<fusion-version>-cdh<version>/lib/wd-netty-all.jar
    
    jar -uf /opt/cloudera/parcels/CDH-<version>.cdh<version>/jars/spark-assembly-<version>-cdh<version>-hadoop<version>-cdh<version>.jar com/** org/** META-INF/**
  2. You now have both the original Spark assembly jar (with the extension “.original”) and a version with the Fusion client libraries available in it. The updated version needs to be made available on each node in the cluster in the /opt/cloudera/parcels/CDH-5.9.1-1.cdh5.9.1.p0.4/jars/ directory.

  3. If you need to revert to the original Spark assembly jar, simply copy it back in place on each node in the cluster.

Spark Assembly Upgrade

The following example covers how you may upgrade the Spark Assembly as part of a Fusion upgrade. This example uses CDH 5.11, although it can be applied generically:

# Create staging path for client and spark assembly
mkdir -p /tmp/spark_assembly/assembly
# Copy existing Spark assembly to work on
cp /opt/cloudera/parcels/CDH/jars/spark-assembly-*.jar /tmp/spark_assembly/assembly/
# Collect file list for purging, sanitise the list as follows
# * List jar files. Do not list symlinks
# * Exclude directory entries which end with a '/'
# * Sort the list
# * Ensure output is unique
# * Store in file
find /opt/cloudera/parcels/FUSION/lib -name '*.jar' -type f -exec jar tf {} \; | grep -Ev '/$' | sort | uniq > /tmp/spark_assembly/old_client_classes.txt
# Purge assembly copy
xargs zip -d /tmp/spark_assembly/assembly/spark-assembly-*.jar < /tmp/spark_assembly/old_client_classes.txt

The resulting spark-assembly is now purged and requires one of two actions:

  • If WANdisco Fusion is being removed, distribute the new assembly to all hosts.

  • If Fusion is being upgraded, retain this jar for the moment and use it within the assembly packaging process for the new client.

3.5.15. Spark 2

Spark 2 comes with significant performance improvements at the cost of incompatibility with Spark (1). The installation of Spark 2 is more straight forward but there is one known issue concerning the need to restart the Spark 2 service during a silent installation. Without a restart, configuration changes will not be picked up.

Spark 2 on HDP 3.x

If you are using HDP 3.x, an additional step is required after deploying Fusion, before running any Spark jobs.

  1. Go to Spark 2 service

  2. Click ConfigsAdvanced spark2-defaults

  3. Change spark.sql.hive.metastore.jars from /usr/hdp/current/spark2-client/standalone-metastore/ to /usr/hdp/current/spark2-client/standalone-metastore/:/opt/wandisco/fusion/client/lib/*

If Spark 2 is installed after WANdisco Fusion you will need to manually symlink the WANdisco Fusion client libraries.

Not required if using the NameNode Proxy
If the Hadoop cluster is using the NameNode Proxy rather than the Fusion Client, then these symlinks are not required.

For HDP, create the 3 symlinks as follows:

ln -s /opt/wandisco/fusion/client/lib/* /usr/hdp/current/spark2-client/jars
ln -s /opt/wandisco/fusion/client/lib/* /usr/hdp/current/spark2-historyserver/jars
ln -s /opt/wandisco/fusion/client/lib/* /usr/hdp/current/spark2-thriftserver/jars

Cloudera will automatically handle the creation of symlinks for managed clusters. However if you are using unmanaged clusters you will need to create the symlinks using the following command:

ln -s /opt/wandisco/fusion/client/lib/* /opt/cloudera/parcels/SPARK2/lib/spark2/jars/

3.5.16. HBase (Cold Back-up mode)

It’s possible to run HBase in a cold-back-up mode across multiple data centers using WANdisco Fusion, so that in the event of the active HBase node going down, you can bring up the HBase cluster in another data centre, etc. However, there will be unavoidable and considerable inconsistency between the lost node and the awakened replica. The following procedure should make it possible to overcome corruption problems enough to start running HBase again, however, since the damage dealt to underlying filesystem might be arbitrary, it’s impossible to account for all possible corruptions.

Requirements

For HBase to run with WANdisco Fusion, the following directories need to be created and permissioned, as shown below:

Table 1. Requirements

platform

path

permission

CDH5.x

/user/hbase

hbase:hbase

HDP2.x

/hbase /user/hbase

Known problem: permissions error blocks HBase repair.

Error example:

2016-09-22 17:14:43,617 WARN  [main] util.HBaseFsck: Got AccessControlException when preCheckPermission

org.apache.hadoop.security.AccessControlException: Permission denied: action=WRITE path=hdfs://supp16-vm0.supp:8020/apps/hbase/data/.fusion user=hbase

        at org.apache.hadoop.hbase.util.FSUtils.checkAccess(FSUtils.java:1685)
        at org.apache.hadoop.hbase.util.HBaseFsck.preCheckPermission(HBaseFsck.java:1606)
        at org.apache.hadoop.hbase.util.HBaseFsck.exec(HBaseFsck.java:4223)
        at org.apache.hadoop.hbase.util.HBaseFsck$HBaseFsckTool.run(HBaseFsck.java:4063)
        at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:70)
        at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:84)

You can configure the root path for all .fusion directories associated with Deterministic State Machines (DSMs).

Customizable DSM token directories

Dsm Token Directory

Configuration Parameter

LocalFs (HCFS)

fusion.dsmToken.dir

S3

fs.fusion.s3.dsmTokenDir

Swift

fs.fusion.swift.dsmTokenDir

These can be set in the respective configurations to change the location of the .fusion directory. It is important to note that the configuration and same path must be added to all fusion servers in all zones if used.

Procedure

The steps below provide a method of handling a recovery using a cold back-up. Note that multiple HMaster/region servers restarts might be needed for certain steps, since hbck command generally requires master to be up, which may require fixing filesystem-level inconsistencies first.

  1. Delete all recovered.edits directory artifacts from possible log splitting for each table/region. This might not be strictly necessary, but could reduce the numbers of errors observed during startup.

    hdfs dfs -rm /apps/hbase/data/data/default/TestTable/8fdee4924ac36e3f3fa430a68b403889/recovered.edits
  2. Detect and clean up (quarantine) all corrupted HFiles in all tables (including system tables - hbase:meta and hbase:namespace). Sideline option forces hbck to move corrupted HFiles to a special .corrupted directory, which could be examined/cleanup up by admins:

    hbase hbck -checkCorruptHFiles -sidelineCorruptHFiles
  3. Attempt to rebuild corrupted table descriptors based on filesystem information:

    hbase hbck -fixTableOrphans
  4. General recovery step - try to fix assignments, possible region overlaps and region holes in HDFS - just in case:

    hbase hbck -repair
  5. Clean up ZK. This is particularly necessary if hbase:meta or hbase:namespace were messed up (note that exact name of ZK znode is set by cluster admin).

    hbase zkcli rmr /hbase-unsecure

    Final step to correct metadata-related errors.

    hbase hbck -metaonly
    hbase hbck -fixMeta

3.5.17. Apache Phoenix

The Phoenix Query Server provides an alternative means for interaction with Phoenix and HBase. When WANdisco Fusion is installed, the Phoenix query server may fail to start. The following workaround will get it running with Fusion.

  1. Open up phoenix_utils.py, comment out.

    #phoenix_class_path = os.getenv('PHOENIX_LIB_DIR','')

    and set WANdisco Fusion’s classpath instead (using the client jar file as a colon separated string). e.g.

    def setPath():
    PHOENIX_CLIENT_JAR_PATTERN = "phoenix-*-client.jar"
    PHOENIX_THIN_CLIENT_JAR_PATTERN = "phoenix-*-thin-client.jar"
    PHOENIX_QUERYSERVER_JAR_PATTERN = "phoenix-server-*-runnable.jar"
    PHOENIX_TESTS_JAR_PATTERN = "phoenix-core-*-tests*.jar"
    
    # Backward support old env variable PHOENIX_LIB_DIR replaced by PHOENIX_CLASS_PATH
    global phoenix_class_path
    #phoenix_class_path = os.getenv('PHOENIX_LIB_DIR','')
    phoenix_class_path = "/opt/wandisco/fusion/client/lib/fusion-client-hdfs-2.6.7-hdp-2.3.0.jar:/opt/wandisco/fusion/client/lib/fusion-client-common-2.6.7-hdp-2.3.0.jar:/opt/wandisco/fusion/client/lib/fusion-netty-2.6.7-hdp-2.3.0.jar:/opt/wandisco/fusion/client/lib/netty-all-4.0.23.Final.jar:/opt/wandisco/fusion/client/lib/guava-11.0.2.jar:/opt/wandisco/fusion/client/lib/fusion-common-2.6.7-hdp-2.3.0.jar"
    if phoenix_class_path == "":
    phoenix_class_path = os.getenv('PHOENIX_CLASS_PATH','')
  2. Edit: queryserver.py, change the Java construction command to look like the one below by appending the phoenix_class_path to it within the "else" portion of java_home :

if java_home:
    java = os.path.join(java_home, 'bin', 'java')
else:
    java = 'java'

#    " -Xdebug -Xrunjdwp:transport=dt_socket,address=5005,server=y,suspend=n " + \
#    " -XX:+UnlockCommercialFeatures -XX:+FlightRecorder -XX:FlightRecorderOptions=defaultrecording=true,dumponexit=true" + \
java_cmd = '%(java)s -cp ' + hbase_config_path + os.pathsep + phoenix_utils.phoenix_queryserver_jar + os.pathsep + phoenix_utils.phoenix_class_path + \
    " -Dproc_phoenixserver" + \
    " -Dlog4j.configuration=file:" + os.path.join(phoenix_utils.current_dir, "log4j.properties") + \
    " -Dpsql.root.logger=%(root_logger)s" + \
    " -Dpsql.log.dir=%(log_dir)s" + \
    " -Dpsql.log.file=%(log_file)s" + \
    " " + opts + \

3.5.18. Running with Apache HAWQ

In order to get Hawq to work with fusion HDFS client libs there needs to be an update made to the pxf classpath. This can be done in Ambari through the "Advanced pxf-public-classpath" setting adding an entry to the client lib path:

/opt/wandisco/fusion/client/lib/*

3.5.19. Apache Slider

Apache Slider is an application that lets you deploy existing distributed applications on an Apache Hadoop YARN cluster, monitor them and make them larger or smaller as desired - even while the application is running. As these applications run within YARN containers, they are isolated from the rest of the cluster, making Slider an ideal mechanism for running applications that are otherwise incompatible with your Hadoop cluster.

WANdisco Fusion supports the use of Slider via the Slider CLI only, not the Ambari Slider View.

3.5.20. KMS / TDE Encryption and Fusion

TDE (Transparent Data Encryption) is available to enhance their data security. TDE uses Hadoop KMS (Key Management Server) and is typically done using Ranger KMS (in Hortonworks / Ambari installs) or Navigator Key Trustee (Cloudera installs).

In simple terms, a security / encryption key or EEK (encrypted encryption key) is used to encrypt the HDFS data that is physical stored to disk. This encryption occurs within the HDFS client, before the data is transported to the datanode.

The key management server (KMS) centrally holds these EEKs in an encrypted format. ACL (access control lists) defines what users/groups are permitted to do with these keys. This includes creating keys, deleting keys, rolling over (re-encrypting the EEK, not changing the EEK itself), obtaining the EEK, listing the key or keys and so on.

Data encrypted in HDFS is split into encrypted zones. This is the act of defining a path (e.g. /data/warehouse/encrypted1) and specifying which EEK is used to to protect this zone (i.e. the key used to encrypt / decrypt the data). A zone is configured with a single key, but different zones can have different keys. Not all of HDFS needs to be encrypted, only the specific zones (and all sub-directories of that zone) an admin defines are.

A user then needs to be granted appropriate ACL access to a get (specifically the "Get Metadata" and "Decrypt EEK" permissions) the EEK needed, to read / write from the zone.

WANdisco Fusion runs as a HDFS user just like any other user. As such, Fusion will need permissions in order to read / write to an encrypted zone.

Fusion may want to write metadata (consistency check, make consistent and other meta operations), tokens or other items for administrative reasons which may fall under an encrypted zone. Depending on configuration and requirements, the make consistent operation itself will be writing data thus needs access.

Additionally, KMS provides its own Proxyuser implementation which is separate to the HDFS proxyusers. Although this works in the same, defining who is permitted to impersonate another user whilst working with EEKs.

To add complication. The "hdfs" user is typically blacklisted from performing the "Decrypt EEK" function by default. The fact "hdfs" is a superuser means they wield great power in the cluster. That does not mean they are superuser in KMS. As "hdfs" is commonly the default user of choice to use to fix things in HDFS (given the simple fact it overrides permissions), it seems wise to prevent such authority to access EEKs by default. Note: Cloudera also seems to blacklist the group "supergroup" which is the group defined as the superusergroup. That is, any users added to "supergroup" become superusers, however they then also automatically get blacklisted from being able to perform EEK operations.

Configuring Fusion

To configure Fusion for access to encrypted zones, two aspects need to be considered:

  1. The local user that Fusion runs as in HDFS (after kerberos auth_to_local mapping) must be able to access and decrypt EEKs.

  2. Although other users will be performing the requests themselves, the Fusion server will proxy that request. As such, a proxyuser within the KMS configs for the Fusion user must also be provided.

Step-by-step guide

The following items need to be considered within KMS configuration to ensure Fusion has access:

The kms-site configuration (such as Advanced kms-site in Ambari) contains its own auth_to_local type parameter called “hadoop.kms.authentication.kerberos.name.rules”

Ensure that any auth_to_local mapping used for the Fusion principal is also contained here. This can be most easily achieved via simple copy/paste from core-site.xml.

The kms-site configuration (such as Custom kms-site in Ambari) contains proxyuser parameters such as:

hadoop.kms.proxyuser.USERNAME.hosts
hadoop.kms.proxyuser.USERNAME.groups
hadoop.kms.proxyuser.USERNAME.users

Entries should be created for the local Fusion user (after auth_to_local translation) to allow Fusion to proxy/impersonate other users requests. This could be as simple as.

hadoop.kms.proxyuser.USERNAME.hosts=fusion.node1.hostname,fusion.node2.hostname
hadoop.kms.proxyuser.USERNAME.groups=*
hadoop.kms.proxyuser.USERNAME.users =*

In the dbks-site configuration, the parameter hadoop.kms.blacklist.DECRYPT_EEK exists. Ensure this does not contain the username that Fusion uses (after auth_to_local translation).

In the KMS ACLs, such as using Ranger KMS, ensure that the Fusion user (after auth_to_local translation) has "Get Metadata" and "Decrypt EEK" permissions to keys.

This could be granted access to all keys. This will avoid a need to review rules when new keys are added. However, Fusion will only need these permissions to keys that apply to zones that fall within a replicated path. Consideration is needed here based on the user that Fusion has been configured as - either "HDFS" will need access to EEKs, OR the fusion user will need access, OR the supergroup could be given access to EEKs (it is enabled by default on Ambari but disabled on CDH), and then make the Fusion user a member of the supergroup.

Defining replicated paths

Replicated paths must be defined at or below the root of the encrypted zone.

For example, if the encrypted path is /repl1/encrypted, the replicated path/DSM can be defined as e.g.:

  • /repl1/encrypted

  • /repl1/encrypted/subdir

  • /repl1/encrypted/subdir/subdir

  • But not /repl1

This is because Hadoop carries a fixed rule that restricts file and directory renames across encryption zone boundaries. See the Hadoop website for more information. Partial Fusion service will be possible if the path is incorrectly defined, however issues will occur for example during renames or adding user directories. Problems can occur if KMS is on one zone but not another, or if both have KMS, but the encrypted boundary exists in different places.

Troubleshooting

If you do not perform the correct configuration, both local operations (as performed by a client) and/or the replicated actions may fail when the Fusion client is invoked. This should only apply to replicated paths.

So to troubleshoot:

  1. Perform the same command without Fusion (use the -D "fs.hdfs.impl=org.apache.hadoop.hdfs.DistributedFileSystem" parameter if running basic HDFS CLI tests). If clients can read/write encrypted content without Fusion, this points to misconfiguration in the above.

  2. Test with an encrypted but non-replicated directory through Fusion client. If this works, but the replicated directory does not, this suggests issues on the remote cluster.

  3. Look in client side application/service logs for permissions issues. (This may be mapreduce, Hive, HBase Region Server logs etc). This may require debug logging being enabled temporarily.

  4. Search for the path/file under investigation; you are looking for KMS ACL exceptions.

3.5.21. WebWasb

WebHDFS is the implementation of HTTP Rest API for HDFS compatible file systems. WebWasb is simply WebHDFS for the WASB file system.

WebWasb can be installed on the edge node where the ISV applications live. From the edge node, WebWasb can be accessed by referring to localhost and the port 50073.

WebWasb works off of the default file system for the cluster (a specified default container in the default storage account) specified in /etc/hadoop/conf/core-site.xml under the property fs.defaultFS. As an example, if your default storage account is named storage1 and your default container is named container1, you could create a new directory called dir1 within that container by the following WebHDFS command:

curl -i -X PUT http://localhost:50073/WebWasb/webhdfs/v1/dir1?op=MKDIRS

WebWasb commands are case sensitive, so pay specific attention to the casing of "WebWasb" and the operations should all be uppercase.

Azure virtual network

With virtual network integration, Hadoop clusters can be deployed to the same virtual network as your applications so that applications can communicate with Hadoop directly. The benefits include:

  • Direct connectivity of web applications or ISV applications to the nodes of the Hadoop cluster, which enables communication to all ports via various protocols, such as HTTP or Java RPC.

  • Improved performance by not having your traffic go over multiple gateways and load-balancers.

  • Virtual network gives you the ability to process info more securely, and only provide specific endpoints to be accessed publicly.

3.5.22. HttpFS

HttpFS is a server that provides a REST HTTP gateway supporting all HDFS File System operations (read and write), and it is interoperable with the webhdfs REST HTTP API.

If httpFS is installed after WANdisco Fusion, then you will need to manually create the file /etc/hadoop-httpfs/tomcat-deployment/bin/setenv.sh (for HDP) or /var/lib/hadoop-httpfs/tomcat-deployment/bin/setenv.sh (for Cloudera) on the HttpFS node. Then add the following script to the file avoid getting a "ClassNotFound" error.

Example for Hortonworks
# START_FUSION - do not remove this line, or the STOP_FUSION line
(shopt -s nullglob
if [ -d "/opt/wandisco/fusion/client/jars" -a -d "/usr/hdp/current/hadoop-httpfs" ]; then
  for jar in "/opt/wandisco/fusion/client/jars/*"; do
    cp "$jar" "/usr/hdp/current/hadoop-httpfs/webapps/webhdfs/WEB-INF/lib"
  done
fi)
# STOP_FUSION
Example for Cloudera
# START_FUSION - do not remove this line, or the STOP_FUSION line
(shopt -s nullglob
if [ -d "/opt/cloudera/parcels/FUSION/lib" -a -d "/opt/cloudera/parcels/CDH/lib/hadoop-httpfs" ]; then
  for jar in "/opt/cloudera/parcels/FUSION/lib/*"; do
    cp "$jar" "/var/lib/hadoop-httpfs/tomcat-deployment/webapps/webhdfs/WEB-INF/lib"
  done
fi)
# STOP_FUSION

3.5.23. ACL Replication

The ACL replication feature enables replication of ACL changes between zones. By default, ACL changes are only executed on the local zone. In situations where you need ACL rules to be applied across zones, enable the feature by ticking the checkbox.

  • This is a zone setting, so is enabled for the whole zone on which you enable the option.

  • Regardless of the state of this setting, a HDFS client loading FusionHdfs will submit agreements for File ACL changes in HDFS (if it is on a replicated path and not-excluded).

  • While a local zone will always execute a locally generated ACL change, it will only be executed in other zones, if the ACL Replication checkbox is ticked.

To use ACL replication the following need to be set:

  1. The checkbox Enable ACL replication on the ACL Replication - Settings panel needs to be checked (it is checked by default).

  2. If using the fusion:// scheme, add the property fusion.acls.supported to the core-site.xml and set it to true.

3.6. Fusion Features

3.6.1. Marker Files

From 2.12.1.8 onwards, applications can now use Fusion marker files to signal across clusters that files are available in full.

Hadoop does not provide a built-in mechanism for file or directory locking, so applications that require some means of signalling when a set of data has been written in full need to provide some application-specific mechanism for this. In a single Hadoop cluster, applications can use a marker file or directory to indicate to other applications when a particular, application-defined set of files have been completely written. i.e. Application X writes file1, file2, and file3, and only having done so in full creates a file marker. Other applications wait on the arrival of the marker file to determine that file1, file2, and file3 are available in full.

This works for a single cluster because an application performing those creates and writes will not create the marker file until it knows that the other files exist in full, including their content. It is a way of signaling through the file system that content is available.

In an environment where Fusion provides replication, the pattern does not hold across clusters because the marker file may be created before the content for each of the other files has been replicated. An application awaiting the marker file in a non-originating zone may still be able to open the other files for read, and not see their full content, whereas if it waited for the marker file in the originating cluster it would likely see the full content for those files.

The Fusion Markers feature extends the capabilities of Fusion to provide a special designation to a directory whose name ends in the text .fusion_marker. This is called a marker. When created, that directory will not be visible in a non-originating cluster until all content associated with files completed previously in the originating cluster (in or beneath the same directory in which the marker was created) have been replicated and made available in full to the non-originating cluster.

Applications can use this feature to achieve the same functionality that is present in a non-replicated Hadoop file system.

To implement the marker file functionality, files are added to the intended source directory as intended, after these files are all added, the marker file directory must be created to the source path, it must either be created directly to the source path or added to the source path, moving or copying a marker file directory will not work due to Fusion’s internal ordering logic. After the creation of the marker file directory, all other actions on the same path, after the creation of the marker file directory are held until the actions before the marker file directory creation are all finished, once they are finished, the marker file is created and all other actions are now performed.

4. Installation (On-premises)

This section will run through the installation of WANdisco Fusion from the initial steps where we make sure that your existing environment is compatible, through the procedure for installing the necessary components and then finally configuration.

4.1. Pre-requisites Checklist

The following prerequisites checklist applies to both the WANdisco Fusion server and for separate IHC servers.

During the installation, your system’s environment is checked to ensure that it will support WANdisco Fusion the environment checks are intended to catch basic compatibility issues, especially those that may appear during an early evaluation phase.

4.1.1. WANdisco Fusion components

This section describes the components in a WANdisco Fusion deployment.

WANdisco Fusion Components
Figure 10. WANdisco Fusion Components
WANdisco Fusion UI

A separate server that provides administrators with a browser-based management console for each WANdisco Fusion server. This can be installed on the same machine as WANdisco Fusion’s server or on a different machine within your data center.

WANdisco Fusion Server

The WANdisco Fusion Servers receive and coordinate Client requests between the Fusion Zones through the use of the distributed coordination engine (DConE). Incoming data is received by the Fusion Server and written to the local HCFS/HDFS. Multiple Fusion Servers can be deployed in a Zone/Cluster for High Availability purposes.

IHC Server

Inter Hadoop Communication servers handle the traffic that runs between zones or data centers that use different versions of Hadoop. IHC Servers are matched to the version of Hadoop running locally. It’s possible to deploy different numbers of IHC servers at each data center, additional IHC Servers can form part of a High Availability mechanism.

WANdisco Fusion servers don’t need to be collocated with IHC servers
If you deploy using the installer, both the WANdisco Fusion and IHC servers are installed into the same system by default. This configuration is made for convenience, but they can be installed on separate systems. This would be recommended if your servers don’t have the recommended amount of system memory.
WANdisco Fusion Client

Client jar files to be installed on each Hadoop client, such as mappers and reducers that are connected to the cluster. The client is designed to have a minimal memory footprint and impact on CPU utilization.

WANdisco Fusion must not be co-located with HDFS servers (DataNodes, etc)
HDFS’s default block placement policy dictates that if a client is collocated on a DataNode, then that collocated DataNode will receive 1 block of whatever file is being put into HDFS from that client. This means that if the WANdisco Fusion Server (where all transfers go through) is collocated on a DataNode, then all incoming transfers will place 1 block onto that DataNode. In which case the DataNode is likely to consume lots of disk space in a transfer-heavy cluster, potentially forcing the WANdisco Fusion Server to shut down in order to keep the Prevaylers from getting corrupted.

4.1.2. Memory and storage

You deploy WANdisco Fusion/IHC server nodes in proportion to the data traffic between clusters; the more data traffic you need to handle, the more resources you need to put into the WANdisco Fusion server software.

If you plan to locate both the WANdisco Fusion and IHC servers on the same machine then check the collocated Server requirements:

CPUs

Minimum for server deployment: 8 cores
Architecture: 64-bit only.

System memory

There are no special memory requirements, except for the need to support a high throughput of data:
Type: Use ECC RAM
Size: Minimum for WANdisco Fusion server deployment: 48 GB

  • WANdisco Fusion is tested using 16 GB of HEAP, which is a good starting point for a deployment.

  • System memory requirements are matched to the expected cluster size and should take into account the number of files.

  • The more RAM you have, the bigger the supported file system.

Collocation of WANdisco Fusion/IHC servers
16GB of this memory requirement is allocated to the IHC server which is nearly always installed on the WANdisco Fusion server.

Recommended WANdisco Fusion server deployment: 16 GB or more

Storage space

Type: Hadoop operations are storage-heavy and disk-intensive so we strongly recommend that you use enterprise-class Solid State Drives (SSDs).
Size: Recommended: 1 TiB
Minimum: You need at least 250 GiB of disk space for a production environment.

Network Connectivity

Minimum 1Gb Ethernet between local nodes.
WANdisco Fusion has no minimum requirement on network throughput, other than it is sized sufficiently for the expected data volumes.

4.1.3. TCP Port Allocation

Before beginning installation you must have sufficient ports reserved. Below are the default, and recommended, ports.

WANdisco Fusion Server

Fusion port: 6444
Fusion port handles all coordination traffic that manages replication. It needs to be open between all WANdisco Fusion nodes. Nodes that are situated in zones that are external to the data center’s network will require unidirectional access through the firewall.

Fusion HTTP Server Port: 8082
The HTTP Server Port or Application/REST API is used by the WANdisco Fusion application for configuration and reporting, both internally and via REST API. The port needs to be open between all WANdisco Fusion nodes and any systems or scripts that interface with WANdisco Fusion through the REST API.

Fusion HTTPS Server Port: 8084
If SSL is enabled, this port is used for application for configuration and reporting, both internally and via REST API. The port needs to be open between all WANdisco Fusion nodes and any systems or scripts that interface with WANdisco Fusion through the REST API.

Fusion Request port: 8023
Port used by WANdisco Fusion server to communicate with HCFS/HDFS clients. The port is generally only open to the local WANdisco Fusion server, however you must make sure that it is open to edge nodes.

Fusion Server listening port: 8024
Port used by WANdisco Fusion server to listen for connections from remote IHC servers. It is only used in unidirectional mode, but it’s always opened for listening. Remote IHCs connect to this port if the connection can’t be made in the other direction because of a firewall. The SSL configuration for this port is controlled by the same ihc.ssl.enabled property that is used for IHC connections performed from the other side. See Enable SSL for WANdisco Fusion.

IHC ports: 7000-range or 9000-range
7000 range, (the exact port is determined at installation time based on what ports are available), it is used for data transfer between Fusion Server and IHC servers. It must be accessible from all WANdisco Fusion nodes in the replicated system.
9000 range, (the exact port is determined at installation time based on available ports), it is used for an HTTP Server that exposes JMX metrics from the IHC server.

HTTP UI port: 8083
Used to access the WANdisco Fusion UI by end users (requires authentication), it’s also used for inter-UI communication. This port should be accessible from all Fusion servers in the replicated system as well as visible to any part of the network where administrators require UI access.

HTTPS UI port: 8443
If SSL is enabled, this port is used to access the WANdisco Fusion UI by end users (requires authentication), it’s also used for inter-UI communication. This port should be accessible from all Fusion servers in the replicated system as well as visible to any part of the network where administrators require UI access.

Hadoop Platform
Figure 11. Network diagram illustrating basic connections/port arrangement

4.1.4. Software requirements

Operating systems:

Amazon Linux 2
CentOS 6 x86_64
CentOS 7 x86_64
Oracle Linux 6 x86_64
Oracle Linux 7 x86_64
Red Hat Enterprise Linux (RHEL) 6 x86_64
RHEL 7 x86_64
SUSE Linux Enterprise Server (SLES) 11 x86_64
SLES 12 x86_64
Ubuntu 14.04LTS
Ubuntu 16.04LTS

We only support AMD64/Intel64 64-Bit (x86_64) architecture.

Web browsers

We develop and test using the following browsers:

  • Chrome 45 and later

  • Edge 12 and later

  • Firefox 40 and later

  • Safari 9 and later

Other browsers and older versions may be used but bugs may be encountered.

Java

Java JRE 1.8.
Testing and development are done using a minimum of Java JRE 1.8. There is also support for Open JDK 8, which is used in cloud deployments. For other types of deployment we recommend running with Oracle’s Java as it has undergone more testing.

Architecture

64-bit only

Heap size

Set Java Heap Size of to a minimum of 1Gigabytes, or the maximum available memory on your server.
Use a fixed heap size. Give -Xminf and -Xmaxf the same value. Make this as large as your server can support.
Avoid Java defaults. Ensure that garbage collection will run in an orderly manner. Configure NewSize and MaxNewSize Use 1/10 to 1/5 of Max Heap size for JVMs larger than 4GB. Stay deterministic!
When deploying to a cluster, make sure you have exactly the same version of the Java environment on all nodes.

Where’s Java?

Although WANdisco Fusion only requires the Java Runtime Environment (JRE), Cloudera and Hortonworks may install the full Oracle JDK with the high strength encryption package included. This JCE package is a requirement for running Kerberized clusters.
For good measure, remove any older JDK that might be present in /usr/java. Make sure that /usr/java/default and /usr/java/latest point to an instance of java 8 version, your Hadoop manager should install this.

Ensure that you set the JAVA_HOME environment variable for the root user on all nodes. Remember that, on some systems, invoking sudo strips environmental variables, so you may need to add the JAVA_HOME to Sudo’s list of preserved variables.

File descriptor/Maximum number of processes limit

Maximum User Processes and Open Files limits are low by default on some systems. It is possible to check their value with the ulimit or limit command:

ulimit -u && ulimit -n

-u The maximum number of processes available to a single user.
-n The maximum number of open file descriptors.

For optimal performance, we recommend both hard and soft limits values to be set to 64000 or more:

RHEL6 and later: A file /etc/security/limits.d/90-nproc.conf explicitly overrides the settings in security.conf, i.e.:

      # Default limit for number of user's processes to prevent
      # accidental fork bombs.
      # See rhbz #432903 for reasoning.
      * soft nproc 1024 <- Increase this limit or ulimit -u will be reset to 1024

Ambari and Cloudera manager will set various ulimit entries, you must ensure hard and soft limits are set to 64000 or higher. Check with the ulimit or limit command. If the limit is exceeded the JVM will throw an error: java.lang.OutOfMemoryError: unable to create new native thread.

Additional requirements

iptables
Use the following procedure to temporarily disable iptables, during installation:

RHEL 6

  1. Turn off with

    $ sudo chkconfig iptables off
  2. Reboot the system.

  3. On completing installation, re-enable with

    $ sudo chkconfig iptables on

RHEL 7

  1. Turn off with

    $ sudo systemctl disable firewalld
  2. Reboot the system.

  3. On completing installation, re-enable with

    $ sudo systemctl enable firewalld

Comment out requiretty in /etc/sudoers
The installer’s use of sudo won’t work with some linux distributions (CentOS where /etc/sudoer sets enables requiretty, where sudo can only be invoked from a logged in terminal session, not through cron or a bash script. When enabled the installer will fail with an error:

execution refused with "sorry, you must have a tty to run sudo" message
Ensure that requiretty is commented out:
# Defaults	requiretty
SSL encryption
Basics

WANdisco Fusion supports SSL for any or all of the three channels of communication: Fusion Server - Fusion Server, Fusion Server - Fusion Client, and Fusion Server - IHC Server.

keystore

A keystore (containing a private key / certificate chain) is used by an SSL server to encrypt the communication and create digital signatures.

truststore

A truststore is used by an SSL client for validating certificates sent by other servers. It simply contains certificates that are considered "trusted". For convenience you can use the same file as both the keystore and the truststore, you can also use the same file for multiple processes.

Enabling SSL

You can enable SSL through the SSL Settings screen by selecting a suitable Fusion HTTP Policy Type - Enable SSL for the WANdisco Fusion Server / IHC Server. It is also possible to enable SSL through a manual edit of the application.properties file.

Updating SSL
Updating fusion-ui-server truststore for fusion-server SSL cert

If you add/update the fusion-server SSL certificate then you must update the fusion-ui-server trust store with that certificate too, otherwise fusion-ui-server will be unable to communicate with fusion-server.

This is important to consider during installation, when you can’t change the fusion-ui-server truststore from the default JVM truststore in the UI — this will block installation if you use a certificate that isn’t signed by a CA in the default truststore.
Workaround

Use one of the following workarounds:

  • Wait until after completing basic installation before configuring SSL for the fusion-server.

  • Add the fusion-server cert to the default JVM truststore, usually jssecerts or cacerts (in that order of preference)

    • This change may be reverted post install if a custom truststore is configured via the UI.

  • Change the default JVM truststore in the startup script, and then restart the Fusion UI server. This can be achieved by adding the following extra Java arguments to the 'JAVA_ARGS' string in the file /opt/wandisco/fusion-ui-server/lib/init-functions.sh:

    -Djavax.net.ssl.trustStore=/path/to/truststore.jks -Djavax.net.ssl.trustStoreType=jks -Djavax.net.ssl.trustStorePassword=<password>
    • This should be considered a temporary configuration for install only, and should be rolled back post install.

    • The truststore pointed to needs to include the managers CA, for example AWS CA if you are using S3.

  • Add the value ui.truststore.option=DISABLED into the ui.properties file and then restart Fusion UI server.

    • This configuration will turn off trust completely, and should be rectified post install.

Disabling low strength encryption ciphers

Transport Layer Security (TLS) and its predecessor, Secure Socket Layer (SSL) are widely adopted protocols that are used transfer of data between the client and the server through authentication and encryption and integrity.

Recent research has indicated that some of the cipher systems that are commonly used in these protocols do not offer the level of security that was previously thought.

In order to stop WANdisco Fusion from using the disavowed ciphers (DES, 3DES, and RC4), use the following procedure on each node where the Fusion service runs:

  1. Confirm JRE_HOME/lib/security/java.security allows override of security properties, which requires security.overridePropertiesFile=true

  2. As root user:

    mkdir /etc/wandisco/fusion/security
    chown hdfs:hadoop /etc/wandisco/fusion/security
  3. As hdfs user:

    cd /etc/wandisco/fusion/security
    echo "jdk.tls.disabledAlgorithms=SSLv3, DES, DESede, RC4" >> /etc/wandisco/fusion/security/fusion.security
  4. As root user:

    cd /etc/init.d
    1. Edit the fusion-server file to add

      -Djava.security.properties=/etc/wandisco/fusion/security/fusion.security

      to the JVM_ARG property.

    2. Edit the fusion-ihc-server-xxx file to add

      -Djava.security.properties=/etc/wandisco/fusion/security/fusion.security

      to the JVM_ARG property.

      cd /opt/wandisco/fusion-ui-server/lib
    3. Edit the init-functions.sh file to add

      -Djava.security.properties=/etc/wandisco/fusion/security/fusion.security

      to the JAVA_ARGS property.

  5. Restart the fusion server, ui server and IHC server.

4.1.5. Supported versions

Please see the release notes for your specific WANdisco Fusion version for the supported Hadoop distributions and versions.

4.1.6. Supported plugins

WANdisco Fusion works with WANdisco’s family of plugins which can be found in the Fusion Plugin Compatibility Table.

4.1.7. Supported applications

Erasure coding - known issue
Fusion will not work for directories which have Erasure coding enabled

Supported Big Data applications may be noted here, as we complete testing:

Application:

Version Supported:

Tested with:

Syncsort DMX-h:

8.2.4.

See the Knowledge base

4.2. Deployment planning

4.2.1. Licensing

WANdisco Fusion includes a licensing model that can limit operation based on time, the number of nodes and the volume of data under replication. WANdisco generates a license file matched to your agreed usage model. You need to renew your license if you exceeds these limits or if your license period ends. See License renewals.

License Limits

When your license limits are exceeded, WANdisco Fusion will operate in a limited manner, but allows you to apply a new license to bring the system back to full operation. Once a license is no longer valid:

  • Write operations to replicated locations are blocked

  • Warnings and notifications related to the license expiry are delivered to the administrator

  • Replication of data will no longer occur

  • Consistency checks and make consistent operations are not allowed

  • Operations for adding replication rules will be denied

Each different type of license has different limits.

Evaluation license

To simplify the process of pre-deployment testing, WANdisco Fusion is supplied with an evaluation license (also known as a trial license). This type of license imposes limits:

Time limit

No. fusion servers

No. of Zones

Replicated Data

Plugins

Specified IPs

30 days

1-2

1-2

1TB

No

No

Production license

Customers entering production need a production license file for each node. These license files are tied to the node’s IP address. In the event that a node needs to be moved to a new server with a different IP address customers should contact WANdisco’s support team and request that a new license be generated. Production licenses can be set to expire or they can be perpetual.

Time limit

No. fusion servers

No. of Zones

Replicated Data

Plugins

Specified IPs

variable (default: 1 year)

variable (default: 20)

variable (default: 10)

variable (default: 20TB)

Yes

Yes

Unlimited license

For large deployments, Unlimited licenses are available, for which there are no usage limits.

4.2.2. Final Preparations

We’ll now look at what you should know and do as you begin the installation.

Time requirements

The time required to complete a deployment of WANdisco Fusion will in part be based on its size, larger deployments with more nodes and more complex replication rules will take correspondingly more time to set up. Use the guide below to help you plan for deployments.

  • Run through this document and create a checklist of your requirements. (1-2 hours).

  • Complete the WANdisco Fusion installation (about 20 minutes per node, or 1 hour for a test deployment).

  • Complete client installations and complete basic tests (1-2 hours).

Of course, this is a guideline to help you plan your deployment. You should think ahead and determine if there are additional steps or requirements introduced by your organization’s specific needs.

Network requirements
  • See the Pre-requisites Checklist - TCP Port Allocation for a list of the TCP ports that need to be open for WANdisco Fusion.

  • WANdisco Fusion does not require that reverse DNS is set up but it is vital that all nodes can be resolved from all zones.

4.2.3. Security

Requirements for Kerberos

If you are running Kerberos on your cluster you should consider the following requirements:

  • Kerberos is already installed and running on your cluster

  • Fusion-Server is configured for Kerberos as described in the Kerberos section.

  • Kerberos Configuration before starting the installation.

For information about running Fusion with Kerberos, read this guide’s chapter on Kerberos.

Warning about mixed Kerberized / Non-Kerberized zones
In deployments that mix kerberized and non-kerberized zones it’s possible that permission errors will occur because the different zones don’t share the same underlying system superusers. In this scenario you would need to ensure that the superuser for each zone is created on the other zones.

For example, if you connect a Zone that runs CDH, which has superuser 'hdfs' with a zone running MapR, which has superuser 'mapr', you would need to create the user 'hdfs' on the MapR zone and 'mapr' on the CDH zone.

Manual Kerberos configuration

See the Knowledge base for instructions on setting up manual Kerberos settings. You only need these in special cases as the steps have been handled by the installer. See Manual Updates for WANdisco Fusion UI Configuration.

Instructions on setting up auth-to-local permissions, mapping a Kerberos principal onto a local system user. See the Knowledge base article - Setting up Auth-to-local.

4.2.4. Clean Environment

Before you start the installation you must ensure that there are no existing WANdisco Fusion installations or WANdisco Fusion components installed on your elected machines. If you are about to upgrade to a new version of WANdisco Fusion you must first see the Uninstall chapter.

Ensure HADOOP_HOME is set in the environment
Where the hadoop command isn’t in the standard system path, administrators must ensure that the HADOOP_HOME environment variable is set for the root user and the user WANdisco Fusion will run as, typically hdfs. When set, HADOOP_HOME must be the parent of the bin directory into which the Hadoop scripts are installed. Example: if the hadoop command is:
/opt/hadoop-2.6.0-cdh5.4.0/bin/hadoop

then HADOOP_HOME must be set to

/opt/hadoop-2.6.0-cdh5.4.0/.

4.2.5. Installer File

You need to match the WANdisco Fusion installer file to each data center’s version of Hadoop. Installing the wrong version of WANdisco Fusion will result in the IHC servers being misconfigured.

Why installation requires root user

Fusion core and Fusion UI packages are installed using root permissions, using the RPM tool (or equivalent for .deb packages). RPM requires root to run - hence the need for the permissions. The main requirement for running with root is the need for the installer to create the directory structure for WANdisco Fusion components, e.g.

  • application/operation = /opt/wandisco/

  • configuration = /etc/wandisco/

  • logging = /var/log/fusion/

  • running = /var/run/fusion/

Once all files are put into place, the permission and ownership is set the specific fusion user. After the installation of the artifacts, root is not used and the Fusion processes themselves are run as a specific Fusion user (default is "hdfs").

List the included files in the installer package by running the above list command, e.g.

  ./fusion-ui-server-<distro>_<package-type>_installer.sh --list

This will generate a list of included package files.

MapR availability

The MapR versions of Hadoop have been removed from the trial version of WANdisco Fusion in order to reduce the size of the installer for most prospective customers. These versions are run by a small minority of customers, while their presence nearly doubled the size of the installer package. Contact WANdisco if you need to evaluate WANdisco Fusion for MapR.

4.2.6. License File

After completing an evaluation deployment, you will need to contact WANdisco about getting a license file for moving your deployment into production.

4.3. Installation of WANdisco Fusion

The installation section covers all the steps in getting WANdisco Fusion deployed into your clusters and/or cloud infrastructure.

4.3.1. Installation Steps

These steps apply to both Cloudera and Ambari platforms, differences between platforms are highlighted during the procedure.

Changes to installation journey with Fusion 2.14

As of 2.14, the installation of the Fusion Client and changes to the cluster core-site are now performed after the installation journey. This allows the user to choose when to deploy the Fusion configuration into their Cluster. Due to this change, it is possible to install and induct all Fusion nodes in a Zone/Cluster before deploying the Cluster configuration.

Known issue for fusion.replicated.dir.exchange property (WD-FUI-8124)
When upgrading/reinstalling WANdisco Fusion to 2.14.0, there is a known issue whereby the fusion.replicated.dir.exchange property in the core-site will be removed if it already exists during the final step of the UI installation journey. Please see the enablement of Fusion Replication exchange directory during the post-installation steps for the method to re-enable this property.

There is an alternative option for installing Fusion, this is detailed in the Silent Installation section whereby installation and configuration can be completed programmatically.

Installation via sudo-restricted non-root user

In some deployments, it may not be permitted to complete the installation using a root user. It should be possible to complete an installation with a limited set of sudo commands.

On the Fusion server, the user will need to be able to run the following command:

sudo /path/to/fusion-ui-server-<distro>_<package-type>_installer.sh

From then on, the UI installer runs as the user-specified user.

Note that on the management server for a cluster (i.e. Ambari Server, Cloudera Manager), the user will need to perform privileged commands.

Examples

sudo tar -xf <fusion-stack>.tar.gz

sudo ambari-server restart

sudo chown cloudera-scm:cloudera-scm <FUSION-version>-<cdh-version>.parcel

Also, potentially sudo cp if the stack/parcel cannot be directly downloaded to the correct directory.

Installing NameNode Proxy on Ubuntu with HDP

Additional files are required if installing the NameNode Proxy on Ubuntu with HDP.

  1. Run the Fusion installer script as described in the following section.

    1. Fusion will install but the NameNode Proxy installation will fail.

  2. From customer.wandisco.com, download the NameNode Proxy plugin debian package.

  3. Run dpkg --install <nn-plugin.deb>.

  4. Confirm installation by looking in /opt/wandisco/fusion/plugins.

Initial installation via the terminal

Use the following steps to complete the initial installation on the command line interface (CLI) using the Fusion installer file. The administrator will need to enter some configuration details when prompted, these will be outlined below. Once the initial settings are entered through the terminal session, the installation is then completed via a web browser.

The screenshots shown in this section are from an Ambari/Hortonworks installation on CentOS 7 so there may be slight differences compared with your environment (e.g. the platform selection list).

Please note the following assumptions before beginning:

  1. Open a terminal session on your first installation server.

  2. Navigate to the installation file and make it executable.

    chmod u+x fusion-ui-server-<distro>_<package-type>_installer.sh
  3. Execute the file.

    ./fusion-ui-server-<distro>_<package-type>_installer.sh
  4. The installer will now start.

    Verifying archive integrity... All good.
    Uncompressing WANdisco Fusion..............................
    
        ::   ::  ::     #     #   ##    ####  ######   #   #####   #####   #####
       :::: :::: :::    #     #  #  #  ##  ## #     #  #  #     # #     # #     #
      ::::::::::: :::   #  #  # #    # #    # #     #  #  #       #       #     #
     ::::::::::::: :::  # # # # #    # #    # #     #  #   #####  #       #     #
      ::::::::::: :::   # # # # #    # #    # #     #  #        # #       #     #
       :::: :::: :::    ##   ##  #  ## #    # #     #  #  #     # #     # #     #
        ::   ::  ::     #     #   ## # #    # ######   #   #####   #####   #####
    
    Welcome to the WANdisco Fusion installation
    
    You are about to install WANdisco Fusion version 2.14
    
    Do you want to continue with the installation? (Y/n)

    The installer will perform an integrity check, confirm the product version that will be installed, then invite you to continue. Enter Y (or just Enter) to continue the installation.

  5. The installer checks that both Perl and Java are installed on the system.

    Checking prerequisites:
    
    Checking for perl: OK
    Checking for java: OK

    See the Pre-requisites Checklist - Java section for more information about these requirements.

  6. Select the platform version and type from the list of supported platforms. The example given below is from a Hortonworks installation (a Cloudera installation would list CDH versions to choose from).

    Please specify the appropriate platform from the list below:
    
    [0] hdp-2.5.x
    [1] hdp-2.6.0/hdp-2.6.1
    [2] hdp-2.6.2
    [3] hdp-2.6.3
    [4] hdp-2.6.4
    [5] hdp-2.6.5
    [6] hdp-3.0.0
    [7] hdp-3.1.0
    
    Which WANdisco Fusion platform do you wish to use?

    Please ensure you select the correct version for your platform, as common functionality may be impeded without the correct underlying packages.

  7. Next, set the system user for running the Fusion application.

    We strongly advise against running WANdisco Fusion as the root user.
    
    For default HDFS setups, the user should be set to 'hdfs'. However, you should choose a user appropriate for running HDFS commands on your system.
    
    Which user should WANdisco Fusion run as? [hdfs]
    Checking 'hdfs' ...
     ... 'hdfs' found.

    The installer does a search for the commonly used account, displaying this as the default option, e.g. [hdfs].

    MapR requirements - Superuser

    If you install into a MapR cluster then you need to assign the MapR superuser system account/group mapr. This is required if you need to run WANdisco Fusion using the fusion:/// URI.

    See the requirements for MapR impersonation for further details.

  8. Check the summary to confirm that your chosen settings are appropriate:

    Installing with the following settings:
    
    Installation Prefix:                /opt/wandisco
    User and Group:                     hdfs:hadoop
    Hostname:                           <fusion.hostname>
    WANdisco Fusion Admin UI Listening on:    0.0.0.0:8083
    WANdisco Fusion Admin UI Minimum Memory:  128
    WANdisco Fusion Admin UI Maximum memory:  512
    Platform:                           <selected platform and version>
    
    Do you want to continue with the installation? (Y/n)

    If these settings are correct then enter "Y" (or just Enter) to complete the installation of the Fusion server.

  9. The Fusion packages will now be installed.

    Installing <your selected packages> server packages:
      <your selected server package> ... Done
      <your selected ihc-server package> ... Done
    Installing plugin packages:
      <any selected plugin packages> ... Done
    Installing fusion-ui-server package:
      <your selected ui-server package> ... Done
    
    Starting fusion-ui-server:                                 [  OK  ]
    Checking if the GUI is listening on port 8083: .......Done
  10. The WANdisco Fusion server will now start up:

    Please visit <fusion.hostname> to complete installation of WANdisco Fusion
    
    If <fusion.hostname> is internal or not available from your browser, replace this with an externally available address to access it.

    At this point, the WANdisco Fusion server and corresponding IHC server will have been installed. The next step is to configure Fusion through a web browser.

Using Isilon and HDFS

If you have both Isilon and HDFS installed on your cluster, the default replicated filesystem is Isilon. To override this, add the following properties to the ui.properties file:

fusion.hcfs.service.type.override={defaults to false if not in ui.properties}
fusion.hcfs.service.type.override.value={defaults to HCFS if not in ui.properties}

For example, if you want Fusion to replicate HDFS on a cluster which also has Isilon, set:
fusion.hcfs.service.type.override=true

Once the properties are set, restart the Fusion UI service.

service fusion-ui-server restart
Browser-based configuration of WANdisco Fusion

Follow this section to complete the installation by configuring WANdisco Fusion via the user interface on a web browser.

Silent Installation
For large deployments it may be worth considering the Silent Installation option.

To start this process, open a web browser and point it at the URL provided in the last step of the Initial installation via the terminal section.

  1. On the first Welcome screen you’re asked to choose between creating a New Zone or Add to an existing Zone.

    Installer start
    Figure 12. Welcome

    The desired selection will depend on whether this is the first installation in the Zone or not.

    New Zone - Select this option if this is the first server to be installed with Fusion in the Hadoop cluster.

    Add to an existing Zone - Select this option if this is not the first server to be installed in the Hadoop cluster. After selecting this option, you will need to enter the Fully Qualified Domain Name of the first Fusion server in the Zone, as well as the Fusion UI port (default 8083).

    Known issue - HTTP authentication must be disabled when inducting to a new zone or adding to an existing zone

    Before adding/inducting, disable HTTP authentication via Fusion Kerberos on the Settings page.
    Once the nodes are inducted, HTTP authentication can be enabled again.

    When adding to an existing Zone, many of the options will be automatically filled in, as information is pulled from the existing Fusion server in the Zone.

    High Availability for WANdisco Fusion

    It’s possible to enable High Availability in your WANdisco Fusion Zone by adding additional WANdisco Fusion nodes to a Zone. These additional nodes ensure that in the event of a system outage, there will be sufficient nodes still available to maintain replication. Use the Add to an existing Zone option on the additional nodes for a Zone to achieve this.

  2. Run the environmental validation checks by selecting Validate.

    Environment 1
    Figure 13. Environment

    The page provides high level guidance on what is being validated, for further details on the required pre-requisites, see the Pre-requisites Checklist.

    After clicking Validate, the installer will run through a series of checks of your server’s hardware and software setup and warn you if any of WANdisco Fusion’s pre-requisites are missing.

    Environment 2
    Figure 14. Validation results

    Any element that fails the check should be addressed before you continue the installation.

    Warnings may be ignored for the purposes of completing the installation, especially if only for evaluation purposes and not for production. However, when installing for production, you should address all warnings, or at least take note of them and exercise due care if you continue the installation without resolving and revalidating.

  3. Select and Upload the WANdisco Fusion license key. The installer will perform validation checks against the license, some of the checks include;

    • An evaluation license should have an expiry date.

    • That the local network interfaces contain an IP address that the license key is valid for (unless it’s an unlimited license).

      License
      Figure 15. License validation

      The conditions of your license agreement (e.g. number of Fusion Servers) will be shown in the panel on the left-hand side.

  4. Enter settings for the Fusion Server and IHC components.

    WANdisco Fusion Server
    Figure 16. Fusion Server settings
    1. WANdisco Fusion Server

      Fully Qualified Domain Name / IP

      The full hostname for the server.

      We have detected the following hostname/IP addresses for this machine.

      The installer will try to detect the server’s hostname from its network settings. Additional hostnames will be listed on the drop-down selector.

      Fusion Port

      TCP port used by Fusion for replicated traffic (default 6444). Validation will check that the port is free and that it can be bound to.

      Maximum Java heap size (GB)

      Enter the maximum Java Heap value for the Fusion server. The minimum for production is 16GB but 64GB is recommended.

      Advanced options (optional)
      Only apply these options if you fully understand what they do.
      The following advanced options provide a number of low level configuration settings that may be required for installation into certain environments. The incorrect application of some of these settings could cause serious problems, so for this reason we strongly recommend that you discuss their use with WANdisco’s support team before enabling them.
      Custom Fusion Request Port

      You can provide a custom TCP port for the Fusion Request Port (also known as Fusion Client port). The default value is 8023. The Fusion Server will use this port to listen to incoming requests from Fusion Clients.

      Strict Recovery

      Two advanced options are provided to change the way that the Fusion server responds to a system shutdown whereby the Fusion server was not shutdown cleanly. The default setting is to not enforce a panic event in the logs, when detecting an unsafe shutdown. This is suitable for using the product as part of an evaluation effort. However, when operating in a production environment, you may prefer to enforce the panic event which will stop any attempted restarts to prevent possible corruption to the database.

      • DConE panic if db is dirty

        This option lets you enable the strict recovery option for WANdisco’s replication engine, to ensure that any corruption to its prevayler database doesn’t lead to further problems. When the checkbox is ticked, Fusion will log a panic message whenever Fusion is not properly shutdown, either due to a system or application problem.

      • App panic if db is dirty

        This option lets you enable the strict recovery option for Fusion’s database, to ensure that any corruption to its internal database doesn’t lead to further problems. When the checkbox is ticked, Fusion will log a panic message whenever Fusion is not properly shutdown, either due to a system or application problem.

      Push Threshold
      • Set threshold manually

        When checked, the option to set the Blocksize in Bytes will appear. This is the number of bytes that a Client will write before indicating they are available to transfer. This can always be adjusted post-installation, see Set push threshold manually panel under the Fusion Server section of the Settings tab.

      Chunk Size in Bytes

      The size of the 'chunks' used in file transfer between Zones (in Fusion 2.14, default is 32768).

    2. IHC Server
      Enter the settings for the IHC Server.

      IHC
      Figure 17. IHC Server details
      Maximum Java heap size (GB)

      Enter the Maximum Java Heap value for the Inter-Hadoop Communication (IHC) server. The minimum for production is 16GB.

      IHC network interface

      The hostname for the IHC server.

      We have detected the following hostname/IP addresses for this machine.

      The installer will try to detect the server’s hostname from its network settings. Additional hostnames will be listed on the drop-down selector.

      Don’t use Default route (0.0.0.0) for this address
      Use the actual hostname for an interface that is accessible from the other clusters/zones. Default route is already used by the WANdisco Fusion server on the other side to pick up a proper address for the IHC server at the remote end.
      Advanced Options (optional)
      IHC server binding address

      In the advanced settings you can decide which address the IHC server will bind to. The address is optional, by default the IHC server binds to all interfaces (0.0.0.0), using the port specified in the ihc.server field.

  1. Enter the settings for the new Zone.

    IHC
    Figure 18. Zone information
    Zone Details
    Zone Name

    The name used to identify the Zone in which the server operates. Must be unique to other Zones.

    URI Selection

    The default behavior for WANdisco Fusion is to fix all replication to the Hadoop Distributed File System / (or hdfs:/// URI). Setting the HDFS scheme provides the widest support for Hadoop client applications, since some applications cannot support the available fusion:/// URI. Each option is explained below:

    URI Selection
    Figure 19. URI Selection options
    MapR requirements - URI selection

    MapR needs to use WANdisco Fusion’s native fusion:/// URI, instead of the default hdfs:/// URI.

    Ensure that, during installation, you select the Use Fusion URI with HCFS option.

    Namenode Proxy with HDFS

    This option was introduced in Fusion 2.14.1. The NameNode Proxy with HDFS option provides an alternative to the Fusion client to replicate HDFS client calls.

    Fusion NameNode Proxy removes the need for Fusion client libraries by having Fusion coordinate activities performed against the Hadoop NameNode. This allows HDFS client actions made to the HDFS NameNode to be proxied and passed to the Fusion Server for replication.

    Unlike the other URI Selection options (which all use the Fusion client), the NameNode Proxy does not require a client package install for every HDFS client node nor additional classpath changes for various HDFS supported applications. It simplifies the integration of WANdisco Fusion with Hadoop environments.

    HDFS URI with HDFS

    This option is available for deployments where the Hadoop applications support neither the Fusion URI nor the HCFS standards. Fusion operates entirely within the standard HDFS URI scheme.

    This configuration will not allow paths with the fusion:/// uri to be used; only paths starting with / (or hdfs:///). The underlying file system will be an instance of the HDFS DistributedFileSystem, which will support applications that are not written to the HCFS specification.

    Fusion URI with HCFS

    When selected, you need to use fusion:/// for all data that must be replicated over an instance of the Hadoop Compatible File System.

    If your deployment includes Hadoop applications that are either unable to support the Fusion URI or are not written to the HCFS specification, this option will not work.

    Platforms that must be run with Fusion URI with HCFS:

    Azure

    LocalFS

    UnmanagedBigInsights

    UnmanagedSwift

    UnmanagedGoogle

    UnmanagedS3

    UnmanagedEMR

    MapR

    Fusion URI with HDFS

    This differs from the default in that while the Fusion URI is used to identify data to be replicated, the replication is performed using HDFS itself. This option should be used if you are deploying applications that can support the WANdisco Fusion URI but not the Hadoop Compatible File System.

    Fusion URI and HDFS URIs with HDFS

    This "mixed mode" supports all the replication schemes (fusion:///, hdfs:/// and no scheme) and uses HDFS for the underlying file system, to support applications that aren’t written to the HCFS specification.

    Advanced Options
    Only apply these options if you fully understand what they do.
    The following Advanced Options provide a number of low level configuration settings that may be required for installation into certain environments. The incorrect application of some of these settings could cause serious problems, so for this reason we strongly recommend that you discuss their use with WANdisco’s support team before enabling them.
    wdf2.14.1 installer09
    Figure 20. Node - Advanced options
    HTTP UI Host

    To change the host the UI binds to, enter your UI host or select it from the drop down below.

    HTTP UI External Host

    The address external processes should use to connect to the UI on. This is the address used by, for example, the Jump to node button on the Fusion UI Nodes tab. Depending on your system configuration, this may be different to the internal address used when accessing the node via a web browser.

    UI TrustStore Options

    The next section deals with your installation’s Fusion UI TrustStore settings. Select the appropriate radio button option, then click Update and Validate or Update (if you select Disable TrustStore).

    Use JVM TrustStore

    Select Use JVM TrustStore to install with the default JVM TrustStore, e.g. $JAVA_HOME/jre/lib/security/cacerts.

    TrustStoreOptions
    Figure 21. TrustStore Options - JVM
    Use Custom Trust Store

    Select Use Custom Trust Store to install with your own TrustStore.

    TrustStoreOptions
    Figure 22. TrustStore Options - Custom
    Disable Trust Store

    Select Disable Trust Store to continue without setting up a TrustStore.

    TrustStoreOptions
    Figure 23. TrustStore Options - Disable
    Stop WANdisco Fusion UI Server if unable to load the Trust Store

    Tick the checkbox if you want the Fusion UI Server to halt if it doesn’t find the TrustStore.

    Validation

    The installer will validate your TrustStore selection. The truststore.ks file is checked for READ access and the password is checked. If validation fails, ensure that your TrustStore details are correct and then retry the step.

  1. During installation you must provide credentials for a Fusion Administrator account.

    Fusion Administrator Configuration

    This account will function as a user with all available Fusion privileges. This is in contrast to other user roles that are now available if you deploy using an authentication/authorization mechanism, such as LDAP/Active Directory. See Authorization and authentication for further details about this.

    Manager Configuration
    Figure 24. Security - Fusion Administrator
    Username

    The username for the controlling Fusion Administrator account that will be used to access all areas of the WANdisco Fusion UI.

    User Password

    The installer automatically generates a well formed (random) password that you use, or you may use the Generate new password feature.

    Should you ever need to reset the Fusion Administrator password, see Generate a new password

    Once you have completed the installation, the Fusion Administrator details can be accessed under the Security section of the Settings Screen (link to section).

    Fusion Admin
    Figure 25. Fusion Administrator settings

    Use this screen to update the User Name or associated Password.

  2. Installation is now complete.

    Start WANdisco Fusion

    Click Start WANdisco Fusion to continue. This will start up the Fusion services on the local node.

    Manager Configuration
    Figure 26. Startup

These steps should be followed for each Fusion node to be installed. Once complete, proceed to the Initial configuration section to set up Fusion for your environment.

4.4. LocalFileSystem Installation

The following section covers the installation and setup of WANdisco Fusion deployed over the LocalFileSystem.

4.4.1. Installer-based LocalFileSystem Deployment

The installation requires an administrator to enter details throughout the procedure. Once the initial settings are entered through the terminal session, the deployment to the LocalFileSystem is then completed through a browser.

The procedure to install for LocalFileSystem (LFS) is very similar to the standard installation outlined in the On-premises installation guide. Follow these steps having read the differences highlighted below.

  • Make sure that you use an LFS specific installer, for example fusion-ui-server-localfs_rpm_installer.sh.

  • During the terminal session of the installer, there is an optional option to choose whether to install the Fusion Client on the current node.

  • On step 6 of the UI installer, there is an optional tickbox for entering Kerberos details if it is a requirement for file system access.

  • On step 6 of the UI installer, the Username and Password for the Fusion UI is defined manually.

Additional property required for LocalFileSystem installations

When deploying a LocalFileSystem installation, there is an additional property that will need to be set after the installation is complete.

  1. On the Fusion node(s) installed for LocalFileSystem, add an additional property to the core-site.xml file that exists in two locations:

    /etc/wandisco/fusion/server/core-site.xml
    /etc/wandisco/fusion/ihc/server/localfs-2.7.0/core-site.xml

  2. Add the following entry to the file(s) and save them afterwards:

    <property><name>fusion.check.ownerAndGroup</name><value>false</value></property>

    Ensure that the entry above is entered after the <configuration> and before the </configuration> in the same format as other properties.

  3. Restart the Fusion services on the node(s) once the additional entries have been added.

    service fusion-ihc-server-localfs_2_7_0 restart
    service fusion-server restart
    service fusion-ui-server restart

LocalFileSystem installations do not set this property at install and the outcome is that Consistency Checks return results that have different usernames / groups, but Fusion cannot resolve them. This property allows Fusion to ignore user/group and perform other checks such as object presence, type and size.

LocalFileSystem files do have usernames, but as Fusion runs as a single user, all replicated objects will be owned by this user and not individual users like in HDFS.

4.4.2. Notes on user settings

When using LocalFileSystem, you can only support a single user. This means when you configure the WANdisco Fusion Server’s process owner, that process owner should also be the process owner of the IHC server, the Fusion UI server, and the client user that will be used to perform any puts.

Fusion under LocalFileSystem only supports 1 user
Again, Fusion under LocalFileSystem only supports 1 user (on THAT side; you don’t have to worry about the other DCs). To assist administrators the LocalFS RPM comes with Fusion and Hadoop shell, so that it is possible to run suitable commands from either. E.g.
hadoop fs -ls /
fusion fs -ls /

Using the shell is required for replication.

4.5. Silent installation

The "Silent" installation tools are still under development, although, with a bit of scripting, it should now be possible to automate WANdisco Fusion node installation. The following section looks at the provided tools, in the form of a number of scripts, which automate different parts of the installation process.

Client Installations
The silent installer does not handle the deployment of client stacks/parcels. You must be aware of the following: Stacks/Parcels must be in place before the silent installer is run, this includes restarting/checking for parcels on their respective managers. Failure to do so will leave the HDFS cluster in a state without fusion clients and running with a config that expects them to be there, this can be fixed by reverting service configs if necessary. See Installing Parcels and Stacks.
How to skip client installation

When installing Fusion, using the full silent installation procedure, on any node other than the name node, you will need to set the following flag to skip the installation of the Ambari client, as this has already been installed. e.g.

./silent-installer_full_install.sh --skip-client-install <location of silent-installer.properties file>

4.5.1. Overview

The silent installation process supports two levels: Unattended installation handles just the command line steps of the installation, leaving the web UI-based configuration steps in the hands of an administrator. See unattended installation.

Fully Automated also includes the steps to handle the configuration without the need for user interaction.

4.5.2. Set the environment

There are a number of properties that need to be set up before the installer can be run:

FUSIONUI_USER

User which will run WANdisco Fusion services. This should match the user who runs the hdfs service.

FUSIONUI_GROUP

Group of the user which will run Fusion services. The specified group must be one that FUSIONUI_USER is in.

Check FUSIONUI_USER is in FUSIONUI_GROUP

Verify that your chosen user is in your selected group.

> groups hdfs
hdfs : hdfs hadoop
FUSIONUI_FUSION_BACKEND_CHOICE

Should be one of the supported package names, as per the following list, which includes all options, not all will be available on a single installer:

Check your release notes
Check the release notes for your version of WANdisco Fusion to be sure the packages are supported on your version.
  • cdh-5.11.0:2.6.0-cdh5.11.0

  • cdh-5.12.0:2.6.0-cdh5.12.0

  • cdh-5.13.0:2.6.0-cdh5.13.0

  • cdh-5.14.0:2.6.0-cdh5.14.0

  • cdh-5.15.0:2.6.0-cdh5.15.0

  • cdh-5.16.0:2.6.0-cdh5.16.0

  • cdh-6.0.0:2.6.0-cdh6.0.0

  • cdh-6.1.0:2.6.0-cdh6.1.0

  • cdh-6.2.0:2.6.0-cdh6.2.0

  • emr-5.3.0:2.7.3-amzn-1

  • emr-5.4.0:2.7.3-amzn-1

  • gcs-1.0:2.7.3

  • gcs-1.1:2.7.3

  • gcs-1.2:2.7.3

  • gcs-1.3:2.7.3

  • hdi-3.6:2.7.3.2.6.2.0-147

  • hdp-2.5.0:2.7.3.2.5.0.0-1245

  • hdp-2.6.0:2.7.3.2.6.0.3-8

  • hdp-2.6.2:2.7.3.2.6.2.0-205

  • hdp-2.6.3:2.7.3.2.6.3.0-235

  • hdp-2.6.4:2.7.3.2.6.4.0-91

  • hdp-2.6.5:2.7.3.2.6.5.0-292

  • hdp-3.0.0:3.1.1.3.0.1.0-136

  • hdp-3.1.0:3.1.1.3.1.0.0-78

  • ibm-4.2:2.7.2-IBM-12

  • ibm-4.2.5:2.7.3-IBM-29

  • localfs-2.7.0:2.7.0

  • mapr-5.0.0:2.7.0-mapr-1506

  • asf-2.7.0:2.7.0

  • asf-3.1.0:3.1.0

This mode only automates the initial command line installation step, the configuration steps still need to be handled manually in the browser steps.

4.5.3. Unattended Installation

Use the following command for an unattended installation where an administrator will complete the configuration steps using the browser UI.

sudo FUSIONUI_USER=x FUSIONUI_GROUP=y FUSIONUI_FUSION_BACKEND_CHOICE=z ./fusion-ui-server_rpm_installer.sh

Example

sudo FUSIONUI_USER=hdfs FUSIONUI_GROUP=hadoop FUSIONUI_FUSION_BACKEND_CHOICE=hdp-2.6.0 ./fusion-ui-server_rpm_installer.sh

4.5.4. Fully Automated Installation

This mode is closer to a full "Silent" installation as it handles the configuration steps as well as the installation.
This must be run as the user chosen to run Fusion. This is normally hdfs.

Properties that need to be set:

SILENT_CONFIG_PATH

Path for the environmental variables used in the command-line driven part of the installation. The paths are added to a file called silent_installer_env.sh.

SILENT_PROPERTIES_PATH

Path to silent_installer.properties file. This is a file that will be parsed during the installation, providing all the remaining parameters that are required for getting set up. The template is annotated with information to guide you through making the changes that you’ll need.
Take note that parameters stored in this file will automatically override any default settings in the installer.

FUSIONUI_USER

User which will run Fusion services. This should match the user who runs the hdfs service.

FUSIONUI_GROUP

Group of the user which will run Fusion services. The specified group must be one that FUSIONUI_USER is in.

FUSIONUI_FUSION_BACKEND_CHOICE

Should be one of the supported package names.

FUSIONUI_UI_HOSTNAME

The hostname for the WANdisco Fusion server.

FUSIONUI_UI_PORT

Specify a fusion-ui-server port (default is 8083)

FUSIONUI_TARGET_HOSTNAME

The hostname or IP of the machine hosting the WANdisco Fusion server.

FUSIONUI_TARGET_PORT

The fusion-server port (default is 8082)

FUSIONUI_MEM_LOW

Starting Java Heap value for the WANdisco Fusion server.

FUSIONUI_MEM_HIGH

Maximum Java Heap.

FUSIONUI_UMASK

Sets the default permissions applied to newly created files. The value 022 results in default directory permissions 755 and default file permissions 644. This ensures that the installation will be able to start up/restart.

FUSIONUI_INIT

Sets whether the server will start automatically when the system boots. Set as "1" for yes or "0" for no

Cluster Manager Variables are deprecated
The cluster manager variables are mostly redundant as they generally get set in different processes though they currently remain in the installer code.

FUSIONUI_MANAGER_TYPE
FUSIONUI_MANAGER_HOSTNAME
FUSIONUI_MANAGER_PORT
FUSIONUI_MANAGER_TYPE

"AMBARI", "CLOUDERA", "MAPR" or "UNMANAGED_EMR" and "UNMANAGED_BIGINSIGHTS" for IBM deployments. This setting can still be used but it is generally set at a different point in the installation now.

validation.environment.checks.enabled

Permits the validation checks for environmental

validation.manager.checks.enabled

Note manager validation is currently not available for S3 installs

validation.kerberos.checks.enabled

Note kerberos validation is currently not available for S3 installs

If this part of the installation fails it is possible to re-run the silent_installer part of the installation by running:

/opt/wandisco/fusion-ui-server/scripts/silent_installer_full_install.sh /path/to/silent_installer.properties

Note that available silent installer scripts are located in:

/opt/wandisco/fusion-ui-server/silent_installer_properties/

i.e.

ls -l /opt/wandisco/fusion-ui-server/silent_installer_properties/
-rw-r--r-- 1 root root  2466 Jul 25 14:16 silent_installer_env.sh
-rw-r--r-- 1 root root 14876 Jul 25 14:16 silent_installer.properties
...

4.5.5. Uninstall WANdisco Fusion UI only

This procedure is useful for installations not requiring the Fusion UI:

sudo yum erase -y fusion-ui-server
sudo rm -rf /opt/wandisco/fusion-ui-server /etc/wandisco/fusion/ui

4.5.6. To Uninstall Fusion UI, Fusion Server and Fusion IHC Server (leaving any fusion clients installed):

See the Uninstall Script Usage Section for information on removing Fusion.

4.5.7. Silent Installation files

For every package of WANdisco Fusion there’s both an env.sh and a .properties file. The env.sh sets environment variables that complete the initial command step of an installation. The env.sh also points to a properties file that is used to automate the browser-based portion of the installer. The properties files for the different installation types are provided below:

silent_installer.properties

standard HDFS installation.

s3_silent_installer.properties

properties file for Amazon S3-based installation.

swift_silent_installer.properties

file for Swift-based installation.

4.5.8. Ambari - Install the WANdisco Fusion stack via the API

These steps can be used as part of an automated install. Ensure the Fusion Client stack is in place on the Ambari Server before attempting to install via the API.

Example
ls -l /var/lib/ambari-server/resources/stacks/HDP/2.5/services | grep FUSION

There should be a FUSION directory present in the services directory.

  1. Add the FUSION service.

    curl -u <username>:<password> -H "X-Requested-By: ambari" http://<ambari-server-host>:8080/api/v1/clusters/<cluster-name>/services -d '{"ServiceInfo":{"service_name":"FUSION"}}'
  2. Add the FUSION_CLIENT component.

    curl -u <username>:<password> -H "X-Requested-By: ambari" http://<ambari-server-host>:8080/api/v1/clusters/<cluster-name>/services/FUSION/components/FUSION_CLIENT -X POST
  3. Get a list of the hosts.

    curl -u <username>:<password> -H "X-Requested-By: ambari" http://<ambari-server-host>:8080/api/v1/clusters/<cluster-name>/hosts/
  4. For each of the hosts in the list, add the FUSION_CLIENT component.

    curl -u <username>:<password> -H "X-Requested-By: ambari" http://<ambari-server-host>:8080/api/v1/clusters/<cluster-name>/hosts/<host-name>/host_components/FUSION_CLIENT -X POST
  5. Install the FUSION_CLIENT component

    curl -u <username>:<password> -H "X-Requested-By: ambari" http://<ambari-server-host>:8080/api/v1/clusters/<cluster-name>/services/FUSION/components/FUSION_CLIENT -X PUT -d '{"ServiceComponentInfo":{"state": "INSTALLED"}}'
  6. Make sure the service components are created and the configurations attached by making a GET call, e.g.

    http://<ambari-server-host>:8080/api/v1/clusters/<cluster-name>/services/FUSION

4.5.9. Cloudera - Installing the WANdisco Fusion parcel via the API

The steps for deployment of the Fusion Client in Cloudera via the API are currently in development. This section will be updated as soon as the steps have been fully qualified.

Ensure the Fusion Client parcel is in place on the Cloudera Management Server before attempting to install via the API.

Example
ls -l /opt/cloudera/parcel-repo | grep FUSION

There should be a FUSION .parcel and .sha file in the parcel repository directory.

4.6. Validation

Before moving into production you should always perform completion testing to ensure that Fusion is functioning as expected, with the minimum required levels of performance.

4.6.1. Environment

Ensure…​

  • operation across two data centers

  • there are three servers that can act as Name Nodes

  • there are a minimum of three data nodes

4.6.2. Networking

  • Ping all hosts by name in both clusters from a central server from which ML will be run

  • Ping either the internal or external repo server from all hosts in the both clusters

  • Do a forward and reverse lookup on all hosts of all other hosts

4.6.3. Operating system

  • Perform the command hostname on all hosts in each cluster and verify that the results from the step above match what is returned by DNS and matches what is in all each nodes hosts file.

  • ssh to all hosts as root in each cluster from a central srever from which

4.7. Upgrade

If you wish to perform a major upgrade, please contact WANdisco support.

To upgrade from an earlier version of WANdisco Fusion to 2.14.0, an outline of the steps involved is:

  1. Run a script to gather information on existing configurations

  2. Upgrade Fusion components

  3. Renew configurations

4.8. Uninstall WANdisco Fusion

To avoid issues during the uninstall process, it is advisable to follow the order below when removing Fusion from the cluster:

  1. Remove all Fusion related properties from the cluster core-site and service configurations.

  2. Depending on what is installed, remove the NameNode Proxy or Fusion service from the cluster (either a or b below).

    1. Remove the NameNode Proxy service from the cluster.

      • If Ambari, delete the NameNode Proxy service and manually uninstall the RPM from the NameNode Proxy host(s).

      • If Cloudera, delete the NameNode Proxy service and deactivate/remove the parcel from hosts.

    2. Remove the Fusion service from the cluster.

      • If Ambari, delete the WANdisco Fusion service and the stack.

      • If Cloudera, deactivate and remove the FUSION parcel from hosts. The parcel can then be deleted.

  3. Restart services on the cluster (this step is included in the sections for service removal mentioned above).

  4. If it was installed, remove the Fusion Client from all hosts (not required if the Cloudera Fusion parcel was used).

  5. Uninstall Fusion from the Fusion servers.

Note on Plugins

Please note that this section does not cover the uninstallation of plugins (except for the NameNode Proxy plugin). Please see the relevant plugin documentation for details on how to uninstall them. In general, plugins should always be removed prior to removing Fusion.

4.8.1. Fusion Touchpoints

In this section, all known configuration changes made as part of a Fusion installation will be documented. This will allow a thorough removal/reversion of Fusion related configuration (if required).

After listing the directories and files, there will be a summary at the end of each sub-section detailing what will need to be removed manually. This is purely informational as the corresponding steps to remove these files and directories are covered in the appropriate sections later in the uninstallation guidance.

Fusion servers

The following directories will be created on the Fusion nodes if they did not exist previously.

/opt/wandisco
/etc/wandisco
/var/log/fusion
/var/run/fusion

Service scripts for Fusion will also be created in the following locations:

/etc/init.d/fusion-ihc-server-<distro-version>
/etc/init.d/fusion-server
/etc/init.d/fusion-ui-server

If using the uninstall Fusion script (documented here), the service scripts will be removed. The purge flag can be used when running the uninstall script to automatically remove the directories listed (they will not be removed by default).

Ambari configuration

Publishing changes in the Fusion UI will result in the core-site being updated on the local Ambari manager. The properties added to the core-site will vary depending on the configuration of Fusion. The following properties will be added regardless of configuration:

HDFS

core-site.xml
fs.fusion.underlyingFs
fusion.client.ssl.enabled
fusion.server
fusion.http.authentication.enabled
fusion.http.authorization.enabled

A list of properties that may be added to the core-site can be found in the Fusion Configuration Properties section.

Additional properties added or appended to Ambari service configurations (not relating to core-site):

YARN

Advanced yarn-log4j - appended

log4j.logger.com.wandisco.fusion.client.BypassConfiguration=OFF log4j.logger.com.wandisco.fs.client=OFF

MapReduce2

Advanced mapred-site

mapreduce.application.classpath=:/opt/wandisco/fusion/client/lib/* - appended

Tez

Advanced tez-site

tez.cluster.additional.classpath.prefix=:/opt/wandisco/fusion/client/lib/* - appended

Storm

Custom storm-site

topology.classpath=/opt/wandisco/fusion/client/lib/*

Ambari Metrics

Advanced ams-hbase-env

hbase_classpath_additional=/opt/wandisco/fusion/client/lib/*

Slider

Advanced slider-env: slider-env template - appended

export SLIDER_CLASSPATH_EXTRA=$SLIDER_CLASSPATH_EXTRA:`for i in /opt/wandisco/fusion/client/lib/*;do echo -n "$i:" ; done`

All these properties will need to be removed manually during uninstallation of Fusion. Instructions for this are documented later for Ambari.

NameNode Proxy - Ambari cluster touchpoints

The following directories will be created on the specified Ambari nodes when the NameNode Proxy is installed. Only the parent directories for all relevant files and subdirectories will be listed.

Directories created on the Ambari Server:

/var/lib/ambari-server/resources/mpacks/nnproxy-mpack-<nnproxy-version>
/var/lib/ambari-server/resources/stacks/HDP/<hdp-version>/services/NN_PROXY
/var/lib/ambari-server/resources/common-services/NN_PROXY

All these directories will need to be removed manually after uninstalling NameNode Proxy.

Directories created on nodes with a NameNode Proxy Master installed (if they did not exist previously):

/opt/wandisco
/etc/wandisco
/var/log/fusion
/var/lib/ambari-agent/cache/stacks/HDP/<hdp-version>/services/NN_PROXY

If the NameNode Proxy Master is co-located on a Fusion server, then only the /var/lib/ambari-agent/cache/stacks/HDP/<hdp-version>/services/NN_PROXY will need to be manually removed after uninstalling NameNode Proxy.

If the NameNode Proxy Master is not co-located on a Fusion server, then the /opt/wandisco will also need to be manually removed after uninstallation.

Fusion Client - Ambari cluster touchpoints

The following directories will be created on the specified Ambari nodes when the Fusion Client is installed. Only the parent directories for all relevant files and subdirectories will be listed.

Directories created on the Ambari Server:

/var/lib/ambari-server/resources/stacks/HDP/<hdp-version>/services/FUSION

Directories created on nodes with a Fusion Client installed (if they did not exist previously):

/opt/wandisco
/etc/wandisco
/var/log/fusion
/var/lib/ambari-agent/cache/stacks/HDP/<hdp-version>/services/FUSION

Fusion Client distribution will update the following directories with symlinks to the Fusion Client libraries:

/usr/hdp/<hdp-version>/hadoop/lib
/etc/alternatives
/var/lib/alternatives

Deletion of the Fusion service and removal of the Fusion Client will also remove these directories and files, except for the /var/lib/ambari-server/resources/stacks/HDP/<hdp-version>/services/FUSION directory on the Ambari Server. This will have to be removed manually.

Fusion Client - Ambari services

Fusion Client distribution will update the following directories with symlinks or copies of the Fusion Client libraries (if the directories existed):

Accumulo
/usr/hdp/<hdp-version>/accumulo/lib/ext

Ambari Metrics
/usr/lib/ambari-metrics-collector - Ambari Metrics Collector

Atlas
/usr/hdp/<hdp-version>/atlas/libext

Druid
/usr/hdp/<hdp-version>/druid/extensions/druid-hdfs-storage

Knox
/usr/hdp/<hdp-version>/knox/ext/ranger-knox-plugin-impl

Oozie
/usr/hdp/<hdp-version>/oozie/libext
/usr/hdp/<hdp-version>/oozie/oozie-server/webapps/oozie/WEB-INF/lib - Oozie server

Ranger
/usr/hdp/<hdp-version>/ranger-admin/ews/webapp/WEB-INF/lib

Spark
/usr/hdp/<hdp-version>/livy/jars - Livy Server

Spark2
/usr/hdp/<hdp-version>/spark2/jars
/usr/hdp/<hdp-version>/livy2/jars - Livy2 server

All of these symlinks/copies will be removed during the uninstallation of the Fusion Client.

Additional changes

HttpFS
Client distribution will also create a script on any node that contains a HttpFS server (webhdfs access via the REST HTTP API):

/usr/lib/bigtop-tomcat/bin/setenv.sh

When the HttpFS server is started, this will ensure the Fusion Client libraries are automatically copied to the classpath directory for HttpFS:

/usr/hdp/<hdp-version>/hadoop-httpfs/webapps/webhdfs/WEB-INF/lib

These scripts will need to be edited or removed manually during uninstallation of Fusion. This is documented later in the HttpFS reversion section.

Spark
If Spark 1 is installed, the Spark Assembly jar may need to have been updated with the Fusion Client libraries. See the updating the Spark Assembly JAR section for details of this procedure.

/usr/hdp/<hdp-version>/spark/lib/spark-assembly-<spark-version>.jar

A copy of the original Spark Assembly jar will have been made as part of the update procedure. When uninstalling Fusion, the original will need to be copied back into place. This is documented later in the Spark reversion section.

Cloudera configuration

Publishing changes in the Fusion UI will result in the core-site being updated on the local Cloudera Management Server. The properties added to the core-site will vary depending on the configuration of Fusion. The following properties will be added regardless of configuration:

HDFS

core-site.xml
fs.fusion.underlyingFs
fusion.client.ssl.enabled
fusion.server
fusion.http.authentication.enabled
fusion.http.authorization.enabled

A list of properties that may be added to the core-site can be found in the Fusion Configuration Properties section.

Additional properties added or appended to Cloudera service configurations (not relating to core-site), these are currently added after Step 6 (Security) of the Fusion UI installation journey:

YARN

Gateway Logging Advanced Configuration Snippet (Safety Valve) - appended

log4j.logger.com.wandisco.fusion.client.BypassConfiguration=OFF log4j.logger.com.wandisco.fs.client=OFF

All these properties will need to be removed manually during uninstallation of Fusion. Instructions for this are documented later for Cloudera.

NameNode Proxy - Cloudera cluster touchpoints

The following directories will be created on the specified Cloudera nodes when the NameNode Proxy is installed. Only the parent directories for all relevant files and subdirectories will be listed.

Files added on the Cloudera Management Server for NameNode Proxy distribution:

/opt/cloudera/parcel-repo/NAMENODE_PROXY-<nnproxy-parcel-version>.parcel
/opt/cloudera/parcel-repo/NAMENODE_PROXY-<nnproxy-parcel-version>.parcel.sha
/opt/cloudera/parcel-repo/NAMENODE_PROXY-<nnproxy-parcel-version>.parcel.torrent
/opt/cloudera/csd/NAMENODE_PROXY-CSD.jar

Directories and files added on all Cloudera nodes after NameNode Proxy distribution:

/opt/cloudera/parcels/NAMENODE_PROXY-<nnproxy-parcel-version>
/opt/cloudera/parcel-cache/NAMENODE_PROXY-<nnproxy-parcel-version>.parcel.torrent

Directories created on nodes with a NameNode Proxy Master installed (if they did not exist previously):

/etc/alternatives/live-nn-conf
/var/lib/alternatives/live-nn-conf
/etc/wandisco
/var/log/fusion

The following files will need to be removed manually on the Cloudera Management Server after uninstalling NameNode Proxy:

/opt/cloudera/csd/NAMENODE_PROXY-CSD.jar
/opt/cloudera/parcel-cache/NAMENODE_PROXY-<nnproxy-parcel-version>.parcel.torrent

The rest will be automatically removed during uninstallation.

Fusion Client - Cloudera cluster touchpoints

The following directories and files will be created on the specified Cloudera nodes when the Fusion Client parcel is activated. Only the parent directories for all relevant files and subdirectories will be listed (where applicable).

If Impala is not installed on the cluster, then the Fusion IMPALA related files and directories should not be present.

Files added on the Cloudera Management Server for Client distribution:

/opt/cloudera/parcel-repo/FUSION-<fusion-parcel-version>.parcel
/opt/cloudera/parcel-repo/FUSION-<fusion-parcel-version>.parcel.sha
/opt/cloudera/parcel-repo/FUSION_IMPALA-<impala-parcel-version>.parcel
/opt/cloudera/parcel-repo/FUSION_IMPALA-<impala-parcel-version>.parcel.sha
/opt/cloudera/parcel-cache/FUSION-<fusion-parcel-version>.parcel.torrent
/opt/cloudera/parcel-cache/FUSION_IMPALA-<impala-parcel-version>.parcel.torrent

Directories added on the Cloudera Management Server during Client distribution:

/opt/cloudera/parcels/FUSION-<fusion-parcel-version>
/opt/cloudera/parcels/FUSION_IMPALA-<impala-parcel-version>

All nodes that have a Fusion Client installed will have a directory and contents created within (if it did not exist previously):

/opt/cloudera/parcels/FUSION-<fusion-parcel-version>

Deactivation, removal and deletion of the Fusion parcel, and removal of the Fusion Client will remove these directories and files.

Fusion Client - Cloudera services

Fusion Client distribution will update the following directories with symlinks or copies of the Fusion Client libraries (if the directories existed):

Accumulo
/opt/cloudera/parcels/ACCUMULO-<accumulo-version>/lib/accumulo/lib/ext

Cloudera Management Server (Enterprise)
/usr/share/cmf/lib
/usr/share/cmf/cloudera-navigator-server/jars - Cloudera Navigator Server

Oozie
/var/lib/oozie
/opt/cloudera/parcels/CDH-<cdh-version>/lib/oozie/libtools

Solr
/opt/cloudera/parcels/CDH-<cdh-version>/lib/solr/webapps/solr/WEB-INF/lib

All of these symlinks/copies will be removed during the uninstallation of the Fusion Client.

Additional changes

HttpFS
If running a HttpFS server, the following directory may contain a script that ensures the HttpFS service loads the Fusion Client libraries (see the HttpFS section for information about setting this up):

/var/lib/hadoop-httpfs/tomcat-deployment/bin/setenv.sh

These scripts will need to be edited or removed manually during uninstallation of Fusion. This is documented later in the HttpFS reversion section.

Spark
If Spark 1 is installed, the Spark Assembly jar may need to have been updated with the Fusion Client libraries. See the updating the Spark Assembly JAR section for details of this procedure.

/opt/cloudera/parcels/CDH-<cdh-version>/jars/spark-assembly-<spark-version>.jar

A copy of the original Spark Assembly jar will have been made as part of the update procedure. When uninstalling Fusion, the original will need to be copied back into place. This is documented later in the Spark reversion section.

4.8.2. Remove Fusion properties from core-site and cluster services

In order to remove Fusion configuration from the cluster, the Fusion related properties should be removed from the core-site.xml and cluster services.

  1. For the core-site properties, filter/search for "fusion" in the HDFS config of the cluster.

    1. If using Ambari, these properties will reside in "Custom core-site".

      Uninstall - Ambari properties
      Figure 27. Ambari - Fusion properties
    2. If using Cloudera, these properties will reside in "Cluster-wide Advanced Configuration Snippet (Safety Valve) for core-site.xml".

      Uninstall - Cloudera properties
      Figure 28. Cloudera - Fusion properties
  2. Select to remove all these properties.

  3. In addition to the above task, ensure that the fs.hdfs.impl property is also removed. It should only be present if the HDFS URI with HDFS or Fusion URI and HDFS URIs with HDFS scheme was selected for Fusion. Its removal ensures that the native Hadoop class is used (i.e. org.apache.hadoop.hdfs.DistributedFileSystem).

    1. Ambari

      Uninstall - Ambari fs.hdfs.impl
      Figure 29. Ambari - fs.hdfs.impl
    2. Cloudera

      Uninstall - Cloudera fs.hdfs.impl
      Figure 30. Cloudera - fs.hdfs.impl
  4. Once all properties have been removed, select to Save changes afterwards.

  5. In addition, some of the cluster services will contain properties related to Fusion.

    1. Refer to the Ambari configuration or Cloudera configuration sections for guidance on what properties to remove.

    2. Ensure that any changes to service configurations are saved before continuing.

  6. A restart of services is recommended once the Fusion service is removed. This is part of the documented steps in the remove the Fusion service from the cluster and Remove the NameNode Proxy service from the cluster sections, one will be followed depending on your environment.

Additional step if Spark 1 is installed
This step is not required if the NameNode Proxy was installed.

If Spark 1 service is installed in the cluster, then it is highly likely that the Spark Assembly Jar will have been updated to include the Fusion client libraries. This will need to be reverted as part of the uninstallation of Fusion.

It is worth performing this step before proceeding onto the removal of the WANdisco Fusion service, as cluster services will be restarted as part of those steps.

Please see the Spark 1 for Ambari or Spark 1 for Cloudera notes referenced earlier in the uninstall section for background information.

Example to revert Spark assembly

Create a file (e.g. cluster_hosts) containing a list of all nodes in the cluster (one per line), and run a for loop command such as the example below:

Hortonworks
for i in `cat cluster_hosts`; do echo $i && ssh $i mv /usr/hdp/<hdp-version>/spark/lib/spark-assembly-<spark-version>.BACKUP /usr/hdp/<hdp-version>/spark/lib/spark-assembly-<spark-version>.jar; done
Cloudera
for i in `cat cluster_hosts`; do echo $i && ssh $i mv /opt/cloudera/parcels/CDH-<cdh-version>/jars/spark-assembly-<spark-version>.BACKUP /opt/cloudera/parcels/CDH-<cdh-version>/jars/spark-assembly-<spark-version>.jar; done

This example assumes that the original Spark Assembly Jar is present in the same location with the .BACKUP suffix.

A restart of the Spark 1 service will be required after this, although this can be done as part of the cluster restarts that are performed once the WANdisco Fusion service is removed.

Additional step if HttpFS is installed
This step is not required if the NameNode Proxy was installed.

See the HttpFS for Ambari or HttpFS for Cloudera notes referenced earlier in the uninstall section for background information.

If a HttpFS server is installed on the cluster, then it is likely that it was made compatible for the Fusion client. In order to revert this, a script for Tomcat should be edited to remove references to the Fusion client libraries.

  1. On any node that hosts a HttpFS server, edit the setenv.sh script for the HttpFS server.

    Example for Hortonworks

    vi /usr/lib/bigtop-tomcat/bin/setenv.sh

    Example for Cloudera

    vi /var/lib/hadoop-httpfs/tomcat-deployment/bin/setenv.sh

  2. Delete the references to the Fusion client libraries from the script (see output below).

    Example for Hortonworks
    # START_FUSION - do not remove this line, or the STOP_FUSION line
    (shopt -s nullglob
    if [ -d "/opt/wandisco/fusion/client/jars" -a -d "/usr/hdp/current/hadoop-httpfs" ]; then
      for jar in "/opt/wandisco/fusion/client/jars/*"; do
        cp "$jar" "/usr/hdp/current/hadoop-httpfs/webapps/webhdfs/WEB-INF/lib"
      done
    fi)
    # STOP_FUSION
    Example for Cloudera
    # START_FUSION - do not remove this line, or the STOP_FUSION line
    (shopt -s nullglob
    if [ -d "/opt/cloudera/parcels/FUSION/lib" -a -d "/opt/cloudera/parcels/CDH/lib/hadoop-httpfs" ]; then
      for jar in "/opt/cloudera/parcels/FUSION/lib/*"; do
        cp "$jar" "/var/lib/hadoop-httpfs/tomcat-deployment/webapps/webhdfs/WEB-INF/lib"
      done
    fi)
    # STOP_FUSION
  3. Once the references have been removed, save and quit the file.

A restart of the HttpFS service will be required after this, although this can be done as part of the cluster restarts that are performed once the WANdisco Fusion service is removed.

4.8.3. Remove the NameNode Proxy service from the cluster

Follow the instructions for the relevant distribution to remove NameNode Proxy artefacts from the cluster.

Ambari - delete the NameNode Proxy service and stack

Use these steps if wanting to remove the NameNode Proxy service from the Ambari cluster.

If the existing nameservice was repurposed for the NameNode Proxies (documented here), then these steps should be reversed so that the NameNodes are assigned back to their original nameservice.
  1. Set the fs.defaultFS value back to its original nameservice.

    • This is not required if the value was never changed (e.g. if the existing nameservice was repurposed for the NameNode Proxies).

      1. Ambari UIHDFSConfigsAdvanced core-site

      2. Change the fs.defaultFS value back to the original NameNode nameservice.

  2. Stop and delete the NameNode Proxy service.

    1. Ambari UINameNode ProxyActionsStop

    2. Ambari UINameNode ProxyActionsDelete Service

  3. Uninstall the NameNode Proxy rpm/deb file.

    1. Open a terminal session to the NameNode Proxy host(s).

    2. Remove the rpm or deb file with the following command.

      RHEL

      yum remove -y fusion-nn-proxy-<hadoop-version>.noarch

      Ubuntu

      apt remove -y fusion-nn-proxy-<hadoop-version>.deb

    3. Remove the related directories by running the command below on all NameNode Proxy host(s).

      rm -rf /opt/wandisco/nn-proxy-server
      rm -rf /var/lib/ambari-agent/cache/stacks/HDP/<hdp-version>/services/NN_PROXY
  4. On the Ambari Manager, delete the stack from the filesystem.

    1. Open a terminal session to the Ambari Manager.

    2. Remove the stack and related directories by running the commands below.

      rm -rf /var/lib/ambari-server/resources/common-services/NN_PROXY
      rm -rf /var/lib/ambari-server/resources/stacks/HDP/<hdp-version>/services/NN_PROXY
      rm -rf /var/lib/ambari-server/resources/mpacks/nnproxy-mpack-<nnproxy-version>
    3. To fully remove the NameNode Proxy service option (from the Add Service page), restart the Ambari Server.

      ambari-server restart

  5. Restart services as necessary in Ambari. This will allow the configuration changes to go through.

Cloudera - delete the NameNode Proxy service and parcel

Use these steps if wanting to remove the NameNode Proxy service from the Cloudera cluster.

If the existing nameservice was repurposed for the NameNode Proxies (documented here), then these steps should be reversed so that the NameNodes are assigned back to their original nameservice.
  1. Set the fs.defaultFS value back to its original nameservice.

    • This is not required if the value was never changed (e.g. if the existing nameservice was repurposed for the NameNode Proxies).

      1. Cloudera UIHDFSConfigurationCluster-wide Advanced Configuration Snippet (Safety Valve) for core-site.xml

      2. Change the fs.defaultFS value back to the original NameNode nameservice (or remove the property if it did not exist previously).

  2. Stop and delete the NameNode Proxy service.

    1. Cloudera UINameNode ProxyNameNode Proxy ActionsStop

    2. Cloudera UINameNode ProxyNameNode Proxy ActionsDelete

  3. Remove the NameNode Proxy parcel.

    1. Cloudera UIParcelsNAMENODE_PROXYDeactivate

    2. Cloudera UIParcelsNAMENODE_PROXYRemove from Hosts

    3. Cloudera UIParcelsNAMENODE_PROXYDelete

  4. Clean up related files on the Cloudera Manager server.

    1. Remove the following files on the Cloudera Manager server via a terminal session.

      rm -f /opt/cloudera/csd/NAMENODE_PROXY-CSD.jar
      rm -f /opt/cloudera/parcel-cache/NAMENODE_PROXY-asf-<nnproxy-parcel-version>.parcel.torrent
    2. Restart the Cloudera Manager service to remove the Add Service option for NameNode Proxy.

      service cloudera-scm-server restart

  5. Restart services as necessary in Cloudera. This will allow the configuration changes to go through, and the relevant services will be highlighted in the UI.

4.8.4. Remove the Fusion service and client from the cluster

Follow the instructions for the relevant distribution to remove Fusion artefacts from the cluster.

Ambari - remove the WANdisco Fusion service and stack

Use these steps if wanting to remove the WANdisco Fusion service from the Ambari cluster.

  1. Ambari - Remove the WANdisco Fusion service from the cluster (choose option a or b, option a is recommended).

    1. UI - Delete the WANdisco Fusion service via the Service Actions drop-down in Ambari.

      Ambari - WANdisco Fusion
      Figure 31. Ambari - delete WANdisco Fusion service
    2. API - Delete the WANdisco Fusion service using a curl call to Ambari.

      Example

      curl -su <user>:<password> -H "X-Requested-By: ambari"
      http(s)://<ambari-server>:<ambari-port>/api/v1/clusters/<cluster-name>/services/FUSION -X DELETE

      Replacing the following with your specific information:

      • <user>:<password> - login and password used for Ambari.

      • http(s) - Adjust this prefix depending on whether the Ambari Server is SSL enabled or not.

      • <ambari-server>:<ambari-port> - the URL used to access Ambari UI.

      • <cluster> - this refers to the cluster name, it can be seen at the very top next to the Ambari logo.

  2. Remove the stack from the Ambari server, and restart the Ambari server afterwards:

    Example

    1. rm -f /var/lib/ambari-server/resources/stacks/HDP/<hdp-version>/services/FUSION

    2. ambari-server restart

  3. Restart services as necessary in Ambari. This will allow the configuration changes to go through.

Remove the Fusion Client from all nodes in the Ambari cluster

Removing the WANdisco Fusion service does not automatically remove the Fusion client.

As such, the Fusion client RPM/DEB packages will need to be removed manually from all nodes in the cluster.

  1. Run the command below to obtain the client package name, which can then be used to be remove it:

    RHEL

    yum list installed | grep fusion

    Debian

    apt list --installed | grep fusion

  2. Remove the client from all nodes in the cluster (choose option a or b, b is recommended).

    1. Run the command below on each node in the cluster to remove the client package:

      RHEL

      yum remove -y fusion-hcfs-<PLATFORM-VERSION>-client-hdfs.noarch

      Debian

      apt remove -y fusion-hcfs-<PLATFORM-VERSION>-client-hdfs.deb

    2. Alternatively, create a file (e.g. cluster_hosts) containing a list of all nodes in the cluster (one per line), and run a for loop command such as the example below:

      RHEL
      for i in `cat cluster_hosts`; do echo $i && ssh $i yum remove -y fusion-hcfs-<PLATFORM-VERSION>-client-hdfs.noarch; done
      Debian
      for i in `cat cluster_hosts`; do echo $i && ssh $i apt remove -y fusion-hcfs-<PLATFORM-VERSION>-client-hdfs.deb; done
  3. The following output will be seen on all nodes where the Fusion client was installed:

    WANdisco Fusion Client uninstalled successfully.

Cloudera - deactivate and remove the FUSION parcel/client

Use these steps if wanting to remove the WANdisco Fusion service from the Cloudera cluster. This will also uninstall the Fusion client on all nodes in the cluster.

  1. On the Cloudera UI, click on the Parcels icon.

    Uninstall - Cloudera parcels
    Figure 32. Cloudera - parcels
  2. If the FUSION_IMPALA parcel is installed, first select to Deactivate this.

    Uninstall - Cloudera Fusion Impala deactivate
    Figure 33. Cloudera - Deactivate Fusion Impala
  3. On the pop out, change to Deactivate only.

    Uninstall - Cloudera Fusion Impala deactivate only
    Figure 34. Cloudera - Fusion Impala - Deactivate only
  4. Scroll down to FUSION and click Deactivate.

    Uninstall - Cloudera Fusion deactivate
    Figure 35. Cloudera - Deactivate Fusion
  5. On the pop out, select to Deactivate only. Note that the Remove From Hosts step cannot be performed until services are restarted.

    Uninstall - Cloudera Fusion deactivate only
    Figure 36. Cloudera - Fusion - Deactivate only
  6. Restart services as necessary in Cloudera. This will allow the configuration changes to go through, and the relevant services will be highlighted in the UI.

  7. On the Parcels page, click Remove From Hosts for the FUSION_IMPALA parcel (if installed).

    Uninstall - Cloudera Fusion Impala removal
    Figure 37. Cloudera - Fusion Impala - Remove From Hosts

    Confirm the removal in the pop-up for the FUSION_IMPALA parcel by selecting OK.

  8. On the Parcels page, click Remove From Hosts for the FUSION parcel.

    Uninstall - Cloudera Fusion removal
    Figure 38. Cloudera - Fusion - Remove From Hosts

    Confirm the removal in the pop-up for the FUSION parcel by selecting OK. This action will remove the Fusion client from all nodes in the cluster.

  9. For the FUSION_IMPALA parcel (if installed), select to Delete to delete the parcel from the filesystem on the Cloudera Manager.

    Uninstall - Cloudera Fusion Impala deletion
    Figure 39. Cloudera - Fusion Impala - Delete parcel
  10. For the FUSION parcel, select to Delete to delete the parcel from the filesystem on the Cloudera Manager.

    Uninstall - Cloudera Fusion deletion
    Figure 40. Cloudera - Fusion - Delete parcel
Manual removal of the Fusion Client

This step is only required if the FUSION parcel was not used in the distribution of the Fusion client and the client package was installed manually on the cluster nodes.

To uninstall the Fusion client package from all nodes in the Cloudera cluster, follow the same steps provided in the Ambari section - remove the Fusion Client from all nodes in the cluster.

4.8.5. Uninstall Fusion packages from the Fusion nodes

Follow this guidance for removing the core WANdisco Fusion installation from a server. This must be performed on all Fusion nodes that need to be uninstalled.

Use the following script:

/opt/wandisco/fusion-ui-server/scripts/uninstall.sh

  • The script is placed on the node during the installation process.

  • You must run the script as root or invoke sudo.

Prior to running this script, it is highly recommended that a backup of the Fusion configuration is performed. See the "Backup config/log files" section for details.
Usage

Running the script with -h outputs a list of options for the script.

Usage: ./uninstall.sh [-c] [-l] [-p] [-d]
 -c: Backup config to '$CONFIG_BACKUP_DIR' (default: /tmp/fusion_config_backup).
 -d: Dry run mode. Demonstrates the effect of the uninstall without performing the requested actions.
 -h: This help message.
 -l: Backup logs to '$LOG_BACKUP_DIR' (default: /tmp/fusion_log_backup).
 -p: Purge config, log, data files, etc to leave a cleaned up system.
Backup config/log files

Run the script with the -c option to back up your config and -l to back up WANdisco Fusion logs. The files will be backed up to the following location by default:

/tmp/fusion_config_backup/fusion_configs-YYYYMMDD-HHmmss.tar.gz
/tmp/fusion_log_backup/fusion_logs-YYYYMMDD-HHmmss.tar.gz
Change the default backup directory

You can change the locations that the script uses for these backups by adding the following environmental variables:

CONFIG_BACKUP_DIR=/path/to/config/backup/dir
LOG_BACKUP_DIR=/path/to/log/backup/dir

See the example below for usage of these variables:

CONFIG_BACKUP_DIR=/path/to/config/backup/dir LOG_BACKUP_DIR=/path/to/log/backup/dir /opt/wandisco/fusion-ui-server/scripts/uninstall.sh -c -l

This will backup the config and logs to the directories specified, and uninstall Fusion leaving the related directories intact.

Dry run

Use the -d option to test an uninstallation. This option lets you test the effects of an installation, without any actual file changes being made. Use this option to be sure that your uninstallation will do what you expect.

Default uninstallation

Running the script without using any additional options performs the following default actions:

  1. Stops all WANdisco Fusion related services.

  2. Uninstalls the Fusion, IHC and UI servers.

    Known issue in 2.14.1 - manual removal of NameNode Proxy plugin required

    If the NameNode Proxy was installed with Fusion, then the related plugin must be manually removed on all Fusion nodes before running the uninstall script.

    To do this, run the commands below on all of the Fusion nodes with the plugin installed.

    1. If required, check the plugin is installed:
      RHEL
      yum list installed | grep fusion-nn-plugin

      Debian
      apt list --installed | grep fusion-nn-plugin

    2. Remove the plugin:
      RHEL
      yum remove -y fusion-nn-plugin-hadoop-<distro-version>.noarch

      Debian
      apt remove -y fusion-nn-plugin-hadoop-<distro-version>.noarch

    3. Clean up the related plugin directories afterwards with the following commands:
      rm -rf /opt/wandisco/fusion/plugins/live-nn
      rm -rf /etc/wandisco/fusion/plugins/live-nn
      rm -rf /var/log/fusion/plugins/live-nn

    • It does not remove the Fusion Client, this should be done manually (if it has not already been removed).

If the default options are used, then the Fusion directories will need to manually deleted:

Example

rm -rf /opt/wandisco/fusion /opt/wandisco/fusion-ui-server /etc/wandisco/ /var/run/fusion/ /var/log/fusion/

In this example, the /opt/wandisco parent directory has been left intact. Please ensure that any potential files within are safely backed up before removing this.

Uninstallation with config purge

Running the script with -p will also include the removal of all directories that were created as part of the WANdisco Fusion installation.

A known issue in the product may result in the /opt/wandisco directory being left with some remaining artefacts. Once the script has run, you may see the following message:

INFO: The '/opt/wandisco' directory still contains items and so will be kept.

The remaining artefacts can be deleted afterwards alongside the directory, but ensure that there is nothing else worth keeping beforehand.

It is advisable to use the purge (-p) option in the event that you need to complete a fresh installation, although ensure that the current configuration is backed up (see the "Backup config/log files" section for details).

Please note that the backup option is for recording previous configuration and capturing logs for analysis. It isn’t practical for this option to be used to restore an installation.

4.9. Additional installation options

4.9.1. Installing to a custom location

The WANdisco Fusion installer places files into specific directories that are created if they do not exist, (see Installer File section for default locations). We strongly recommend that you use the default locations as they are better supported and more roundly tested. For deployments where these locations are not permitted, the use of symlinks should be considered.

Example

  ln -s /opt/wandisco /path/to/target_opt_directory
  ln -s /etc/wandisco /path/to/target_etc_directory
  ln -s /var/log/wandisco /path/to/target_log_directory
  ln -s /var/run/wandisco /path/to/target_run_directory

The following RPM relocation feature is also available, although this does have certain limitations (listed below in Pre-requisites). The feature allows installations of WANdisco Fusion to a user-selected location.

Pre-requisites
  • RHEL and derivatives only (SLES not currently supported)

  • Special attention will be required for client installations.

  • Limitation concerning Ambari stack installation

Non-root Ambari agents
Unfortunately the Ambari Stack installer cannot be configured for non-root if you intend to use this RPM relocation feature. Ambari can be configured for non-root Ambari Agents.
FUSION_PREFIX Environmental variable

When running the installer, first set the following environmental variable:

sudo FUSION_PREFIX=<custom-directory> ./fusion-ui-server-hdp_rpm_installer.sh

This will change the installation directory from the default to the one that you provide, e.g.

sudo FUSION_PREFIX=/CustomInstallLocation ./fusion-ui-server-hdp_rpm_installer.sh

The above example would install fusion-ui-server into /CustomInstallLocation/fusion-ui-server. Also, the WANdisco Fusion server and IHC server will be installed under /CustomInstallLocation/fusion/server and /CustomInstallLocation/fusion/ihc/server/ respectively.

If you run with the FUSION_PREFIX, an additional line will appear on the summary screen of the installer:

   ::   ::  ::     #     #   ##    ####  ######   #   #####   #####   #####
  :::: :::: :::    #     #  #  #  ##  ## #     #  #  #     # #     # #     #
 ::::::::::: :::   #  #  # #    # #    # #     #  #  #       #       #     #
::::::::::::: :::  # # # # #    # #    # #     #  #   #####  #       #     #
 ::::::::::: :::   # # # # #    # #    # #     #  #        # #       #     #
  :::: :::: :::    ##   ##  #  ## #    # #     #  #  #     # #     # #     #
   ::   ::  ::     #     #   ## # #    # ######   #   #####   #####   #####

Installing with the following settings:

Installation Prefix:                /CustomInstallLocation
User and Group:                     hdfs:hdfs
Hostname:                           localhost.localdomain
Fusion Admin UI Listening on:       0.0.0.0:8083
Fusion Admin UI Minimum Memory:     128
Fusion Admin UI Maximum memory:     512
Platform:                           hdp-2.6.0

Do you want to continue with the installation? (Y/n)
Alternate method

You can also perform an installation to a custom directory using the following alternative:

Inject the environmental variable:

`export FUSION_PREFIX=<custom-directory>`

Run the installer as per the usual method, i.e.:

./fusion-ui-server-hdp_rpm_installer.sh

The installer will use the provided path for the installation, as described in the main procedure.

Custom location installations - Client Installation

When installing to a custom location, you will need to ensure that your clients are configured with the matching location. You should be able to correctly install clients using the normal procedure, outlined above. See Client Installation, immediately below.

4.9.2. Client installation

If you need to download the Fusion Client after installation is complete, use the links on this screen.

WANdisco Fusion Client configuration
Figure 41. Settings - Client Distribution
Installation with RPM

RPM package location
If you need to find the packages after leaving the installer page with the link, you can find them in your installation directory, here:

/opt/wandisco/fusion-ui-server/ui-client-platform/downloads/

If you are installing the RPMs, download and install the package on each of your nodes.

Installing the client RPM is done in the usual way:

rpm -i <package-name>

Install checks

  • First, we check if we can run hadoop classpath, in order to complete the installation.

  • If we’re unable to run hadoop classpath then we check for HADOOP_HOME and run the Hadoop classpath from that location.

  • If the checks cause the installation to fail, you need to export HADOOP_HOME and set it so that the hadoop binary is available at $HADOOP_HOME/bin/hadoop, e.g.

    export HADOOP_HOME=/opt/hadoop/hadoop
    export HIVE_HOME=/opt/hadoop/hive
    export PATH=$HADOOP_HOME/bin:$HIVE_HOME/bin
Installation with DEB

Debian not supported
Although Ubuntu uses Debian’s packaging system, currently Debian itself is not supported. Note: Hortonworks HDP does not support Debian.

If you are running with an Ubuntu Linux distribution, you need to go through the following procedure for installing the clients using Debian’s DEB package:

DEB package location
If you need to find the packages after leaving the installer page with the link, you can find them in your installation directory, here:

/opt/wandisco/fusion-ui-server/ui-client-platform/downloads/client_package
  1. To install WANdisco Fusion client, download and install the package on each of your nodes.

  2. You can install it using

    sudo dpkg -i /path/to/deb/file

    followed by

    sudo apt-get install -f

    Alternatively, move the DEB file to /var/cache/apt/archives/ and then run

    apt-get install <fusion-client-filename.deb>
Parcel Locations

By default local parcels are stored on the Cloudera Manager Server:/opt/cloudera/parcel-repo. To change this location, follow the instructions in Configuring Server Parcel Settings.

The location can be changed by setting the parcel_dir property in /etc/cloudera-scm-agent/config.ini file of the Cloudera Manager Agent and restart the Cloudera Manager Agent or by following the instructions in Configuring the Host Parcel Directory.

Don’t link to /usr/lib/
The path to the CDH libraries is /opt/cloudera/parcels/CDH/lib instead of the usual /usr/lib. We strongly recommend that you don’t link /usr/lib/ elements to parcel deployed paths, as some scripts distinguish between the two paths.
Replacing earlier parcels?

If you are replacing an existing package that was installed using a parcel, once the new package is activated you should remove the old package through Cloudera Manager. Use the Remove From Host button.

WANdisco Fusion tree

Client installation on ASF

If using an ASF setup, follow the client installation instructions below to install the client on each applicable node. Once complete, distribute the core-site.xml to all nodes in the Hadoop cluster.

  1. To install the Client manually, download the from RPM/DEB packages from the UI.

  2. Install on each applicable node

  3. Distribute the core-site.xml. How you do this depends on your setup:

    • If a core-site.xml file exists in the local config directory (default is /usr/hadoop-<version>/etc/hadoop/core-site.xml) and has an entry for fusion.server then this file should be copied and placed on every node where the Fusion client is installed.

    • If a core-site.xml file exists in the local config directory but has no entry for fusion.server, then this file should be updated with the values from the Fusion Server (generated) core site file. This is located at /etc/wandisco/fusion/server/core-site.xml by default. All <property> elements of this file should be merged into the default core-site.xml file. This file should be copied and placed on every node where the Fusion client is installed.

    • If no core-site.xml file exists in the local config directory, then the generated core site file should be copied and placed on every node where the Fusion client is installed. The generated file is located at /etc/wandisco/fusion/server/core-site.xml by default.

4.9.3. Installer Help

Running the following command on the installer will generate a number of helpful options:

./fusion-ui-server-<distro>_<package-type>_installer.sh --help
Makeself version 2.1.5
 1) Getting help or info about ./fusion-ui-server-<distro>_<package-type>_installer.sh :
  ./fusion-ui-server-hdp_rpm_installer.sh --help   Print this message
  ./fusion-ui-server-hdp_rpm_installer.sh --info   Print embedded info : title, default target directory, embedded script ...
  ./fusion-ui-server-hdp_rpm_installer.sh --lsm    Print embedded lsm entry (or no LSM)
  ./fusion-ui-server-hdp_rpm_installer.sh --list   Print the list of files in the archive
  ./fusion-ui-server-hdp_rpm_installer.sh --check  Checks integrity of the archive

 2) Running ./fusion-ui-server-<distro>_<package-type>_installer.sh :
  ./fusion-ui-server-hdp_rpm_installer.sh [options] [--] [additional arguments to embedded script]
  with following options (in that order)
  --confirm             Ask before running embedded script
  --noexec              Do not run embedded script
  --keep                Do not erase target directory after running
			the embedded script
  --nox11               Do not spawn an xterm
  --nochown             Do not give the extracted files to the current user
  --target NewDirectory Extract in NewDirectory
  --tar arg1 [arg2 ...] Access the contents of the archive through the tar command
  --                    Following arguments will be passed to the embedded script

 3) Environment:
  LOG_FILE              Installer messages will be logged to the specified file

Here is one example of an option in use. A workaround if /tmp directory is "noexec":

Running the installer script will write files to the filesystem’s /tmp directory. If the /tmp directory is mounted with the "noexec" option then the following argument can be used when running the installer:

--target /path/to/executable_directory

Example

sudo ./fusion-ui-server-<distro>_<package-type>_installer.sh --target /opt/wandisco/installation/

The target directory should have the following attributes:

  • No write/execute restrictions.

  • Sufficient storage space.

4.9.4. Installing without the installer

The following procedures covers the hands-on approach to installation and basic setup of a deployment that deploys over the LocalFileSystem. For the vast majority of cases you should use the previous Installer-based LocalFileSystem Deployment procedure.

Don’t do it this way unless you have to.
We provide this example to illustrate how a completely hands-on installation can be performed. We don’t recommend that you use it for a deployment unless you absolutely can’t use the installers. Instead, use it as a reference so that you can see what changes are made by our installer.
Non-HA Local filesystem setup
  1. Start with the regular WANdisco Fusion setup. You can go through either the installation manually or using the installer.

  2. When you select the $user:$group you should pick a master user account that will have complete access to the local directory that you plan to replicate. You can set this manually by modifying etc/wandisco/fusion-env.sh setting FUSION_SERVER_GROUP to $group and FUSION_SERVER_USER to $user.

  3. Next, you’ll need to configure the core-site.xml, typically in /etc/hadoop/conf/, and override “fs.file.impl” to “com.wandisco.fs.client.FusionLocalFs”, “fs.defaultFS” to "file:///", and "fs.fusion.underlyingFs" to "file:///". (Make sure to add the usual Fusion properties as well, such as "fusion.server").

  4. If you are running with fusion URI, (via “fs.fusion.impl”), then you should still set the value to “com.wandisco.fs.client.FusionLocalFs”.

  5. If you are running with Kerberos, then you should override “fusion.handshakeToken.dir” to point to some directory that will exist within the local directory you plan to replicate to/from. You should also ensure that the “fusion.keytab” and “fusion.principal” properties are defined as usual.

  6. Ensure that the local directory you plan to replicate to/from already exists. If not, create it and give it 777 permissions or create a symlink (locally) that will point to the local path you plan to replicate to/from.

  7. For example, if you want to replicate /repl1/ but don’t want to create a directory on your root level, you can create a symlink to repl1 on your root level and point it to wherever you want to actually be your replicated directory. In the case of using NFS, it should be used to point to /mnt/nfs/.

  8. Set-up an NFS.

Be sure to point your replicated directory to your NFS mount, either directly or using a a symlink.

HA local file system setup
  1. Install Fusion UI, Server, IHC, and Client (for LocalFileSystem) on every node you plan to use for HA.

  2. When you select the $user:$group you should pick a master user account that will have complete access to the local directory that you plan to replicate. You can set this manually by modifying /etc/wandisco/fusion-env.sh setting FUSION_SERVER_GROUP to $group and FUSION_SERVER_USER to $user.

  3. Next, you’ll need to configure the core-site.xml, typically in /etc/hadoop/conf/, and override “fs.file.impl” to “com.wandisco.fs.client.FusionLocalFs”, “fs.defaultFS” to "file:///", and “fs.fusion.underlyingFs” to "file:///". (Make sure to add the usual Fusion properties as well, such as "fs.fusion.server").

  4. If you are running with fusion URI, (via “fs.fusion.impl”), then you should still set the value to “com.wandisco.fs.client.FusionLocalFs”.

  5. If you are running with Kerberos, then you should override “fusion.handshakeToken.dir” to point to some directory that will exist within the local directory you plan to replicate to/from. You should also ensure that the “fusion.keytab” and “fusion.principal” properties are defined as usual.

  6. Ensure that the local directory you plan to replicate to/from already exists. If not, create it and give it 777 permissions or create a symlink (locally) that will point to the local path you plan to replicate to/from.

  7. For ex, if you want to replicate /repl1/ but don’t want to create a directory on your root level, you can create a symlink to repl1 on your root level and point it to wherever you want to actually be your replicated directory. In the case of using NFS, it should be used to point to /mnt/nfs/.

  8. Now follow a regular HA set up, making sure that you copy over the core-site.xml and fusion-env.sh everywhere so all HA nodes have the same configuration.

  9. Create the replicated directory (or symlink to it) on every HA node and chmod it to 777.

5. Installation (Cloud)

The following section covers the installation of WANdisco Fusion into a cloud / hybrid-cloud environment.

5.1. Alibaba Installation

WANdisco Fusion can be installed on Alibaba Cloud, enabling you to replicate on-premises data over to Alibaba Cloud.

5.1.1. Installation of WANdisco Fusion for use with Alibaba

Installing WANdisco Fusion for use with Alibaba Cloud follows a very similar process to the standard installation.

The first part of the installation is CLI based. Follow the steps in the Starting the Installation section to do this.
Note - you will need to use an Alibaba specific installer, for example, fusion-ui-server-alibaba_rpm_installer.sh.

The next section of the installer is browser based. For this, follow the steps in the Browser-based configuration section. All steps are the same for Alibaba installation except for some additional fields which need to be completed on page 5 of the installer.

Alibaba
Figure 42. Step 5 - Zone information (Alibaba deployment)
Bucket Name

The name of the storage bucket that will be replicated. Validation checks that this bucket is writable.

Bucket Region Endpoint

This is where your bucket is based. As the bucket name contains no identifying information on where it is located, this information is necessary.

Use Path based access

Select this if you want to use path based access, this is important if your bucket is protect by SSL.

Segment Size

This the smallest block to write.

Buffer Directory

This is where complete objects are stored before being pushed to storage. Appends are not supported and so complete files must always be pushed.

Listing Method

This defines which listing request Fusion makes.

If you are using access key and secret key then additional fields are required. Please see the Use access key and secret key section for more information on this.

As there are no clients to install for this Cloud Platform you can skip step 7 of the installer.

For more information on post-installation configuration see the Configuration for S3 section, and for information about completing cloud or hybrid deployments, see the Cloud Deployment Guide.

5.2. Amazon Installation

This section covers how WANdisco Fusion can be used with the Amazon Web Services (AWS) platform.

This section gives information on installing and using the Fusion installation with EMR/S3.

5.2.1. Prerequisites for AWS

See the Prerequisites Checklist for information about what is required for a WANdisco Fusion set up.

There is also more information on, for example, networking in the Cloud Deployment Guide.

Setting up AWS profiles

IAM roles are the preferred method for S3 authentication, they are specific to the AWS platform.

If you don’t have the correct IAM roles to access your S3 bucket you can use the Access Key and Secret Key credentials. These are configured by running aws configure on the command line. This creates a .aws directory with a credentials file with the relevant keys, under a [default] section.

However, there is also the concept of profiles, and you can store multiple different credentials for different profiles using the CLI command.

CLI

The AWS command line can be installed and configured to access S3 buckets.

aws configure --profile <profilename>

Example

aws --endpoint-url=https://s3-api.us-example.domain.url.net --profile fusion s3 ls s3://vwbucket/repl1/

This creates a new section in the credentials file like so:

[newprofilename]
[nolan]
aws_access_key_id = A******XYZ123ABCRFOA
aws_secret_access_key = 77***********************XZ
SSL with AWS

When using SSL between Fusion nodes, you can create dedicated truststores. However, when connecting WANdisco Fusion to external resources, such as AWS/Cloud nodes, the SSL connection will fail because these external nodes use CA certificates.

When using SSL between Fusion and other nodes, such as cloud object stores, you need to update your truststores to include both the homemade certs and the Root authorities certs.

5.2.2. Limitations and considerations

  • When you stop the Fusion EC2 instances, Fusion data on the EBS storage will remain on the root device and its continued storage will be charged for. However, temporary data in the instance stores will be flushed as they don’t need to persist.

  • If the WANdisco Fusion servers are turned off, then replication to the S3 bucket will stop.

5.2.3. Installing into Amazon S3

Installation of WANdisco Fusion in AWS S3 is documented in a separate S3 Installation section, which covers all S3 compatible platforms.

5.2.4. Installing into Amazon Elastic MapReduce Filesystem (EMRFS)

Installation of WANdisco Fusion in AWS EMRFS follows a very similar process to the standard Fusion installation.

CLI

The first part of the installation is CLI based. Follow the steps in the Initial installation via the terminal section to do this.

You will need to use a EMR specific installer, for example, fusion-ui-server-emr_<OS-version>_installer.<fusion-version>.sh.

Note the difference below when you have reached the specify the appropriate platform option:

Please specify the appropriate platform from the list below:

[0] emr-5.3.0
[1] emr-5.4.0

Which WANdisco Fusion platform do you wish to use?

Select the appropriate platform version for your environment to continue.

UI

The next section of the installer is browser based. For this, follow the steps in the Browser-based configuration of WANdisco Fusion section. All steps are the same for EMR installation except for some additional fields which need to be completed on page 5 of the installer.

AWS EMR Install
Figure 43. AWS EMR install
EMR Information
Bucket Name

The name of the S3 Bucket that will connect to Fusion.

Amazon S3 encryption

Enables Amazon S3 encryption, it can be enabled on a per-bucket level in AWS. Check the box if it is enabled on this S3 bucket. See Amazon’s documentation for more information on this feature.

Use KMS with Amazon S3

Use an established AWS Key Management Server, if enabling, an additional KMS Key ID field will be required. See the Use KMS with Amazon S3 section for more information.

Use access key and secret key

If you are using access key and secret key then additional fields are required. See the Use access key and secret key section below for more information on this. These settings can also be updated on the Settings tab in the UI after installation.

Click Validate after entering the details so that the checks are performed to ensure that Fusion can correctly interact with the S3 Bucket. This may need to be performed after entering Access/Secret key details (if applicable).

Core-Site.xml Information
fs.s3.buffer.dir

The full path to a directory or multiple directories, separated by comma without space, that S3 will use for temporary storage. The install will check that the directory exists and that it will accept writes.

hadoop.tmp.dir

The full path to a one or more directories that Hadoop will use for housekeeping data storage. The installer will check that the directories that you provide exists and is writable. You can enter multiple directories separate by comma without space.

Click Validate after entering the details to check that Fusion can write to the directories specified in the Core-Site.xml Information fields.

Use KMS with Amazon S3
KMS Key ID

This option must be selected if you are deploying your S3 bucket with AWS Key Management Service. Enter your KMS Key ID. This is a unique identifier of the key. This can be an ARN, an alias, or a globally unique identifier. The ID will be added to the JSON string used in the EMR cluster configuration.

5.2.5. Installing the Fusion client on a new Amazon Elastic MapReduce (EMR) cluster

In order to allow Live replication on the EMR cluster, the Fusion client must be installed.

These instructions are for installing the client on a new AWS EMR cluster. This is the recommended approach, even if you already have an EMR cluster set up.

  1. On the Settings tab, go to EMR Client and follow the instructions to configure EMR for Fusion.

    EMR Client - install
    Figure 44. EMR Client - install
  2. Select to Place files, followed by Confirm. This will automatically generate configuration scripts for your AWS cluster and place the scripts into the S3 bucket assigned during the Fusion installation.

    wdfusioninstall

    The folder created to contain the files.

    EMR_config_JSON

    The JSON file containing EMR configuration.

    emrFusionClientScript.sh

    The script Fusion has generated to install clients.

    fusion-hcfs-emr rpm

    (Note: The version number may vary.) This file is used by the emrFusionClientScript.sh script. It contains the Fusion clients.

  3. Proceed to create the Amazon EMR cluster but ensure that Go to advanced options is selected.

    In most cases, you will run with the same settings that would apply without WANdisco Fusion in place, but note the steps required after this.

  4. In Step 1: Software and Steps, ensure to change the following configuration.

    Edit software settings
    Load JSON from S3

    Enter the absolute path to the EMR_config_JSON file in the S3 bucket assigned to Fusion.

    Example

    s3://fusion_install/wdfusioninstall/EMR_config_JSON
  5. In Step 3: General Cluster Settings, ensure to add the following configuration.

    Bootstrap Actions
    Add Bootstrap action

    Select Custom action from the drop-down list, and then Configure and add.

  6. For the Script location field, navigate to the emrFusionClientScript.sh script, generated by Fusion detailed in an earlier step.

    Example

    s3://fusion_install/wdfusioninstall/emrFusionClientScript.sh

    Leave the Optional arguments field empty and select Add.

  7. Finally, click the Create cluster button to complete the AWS setup.

5.2.6. Installing the Fusion client on an existing Amazon Elastic MapReduce (EMR) cluster

Please be aware of the following considerations/limitations when deploying the Fusion client on an existing EMR cluster (rather than a new one).

No autoscaling

This is because installing WANdisco Fusion into an existing cluster will not benefit from AWS’s auto-scaling feature.

Manual core-site change

The configuration changes that you make to the core-site.xml file will not be included in automatically generated cluster nodes. As the cluster automatically grows, you will have to follow up by manually distributing the client configuration changes.

Additional steps when deploying on an existing EMR cluster

Install the Fusion client (the one for EMR) on each node and after scaling, modify the core-site.xml file with the following:

Example

<property>
  <name>fs.fusion.underlyingFs</name>
  <value>s3://YOUR-S3-URL/</value>
</property>
<property>
  <name>fs.fusion.server</name>
  <value>FUSION-HOSTNAME:8023</value>
</property>
<property>
  <name>fs.fusion.impl</name>
  <value>com.wandisco.fs.client.FusionHcfs</value>
</property>
<property>
  <name>fs.AbstractFileSystem.fusion.impl</name>
  <value>com.wandisco.fs.client.FusionAbstractFs</value>
</property>
fs.fusion.underlyingFs

The address of the underlying filesystem. In the case of Elastic MapReduce FS, the fs.defaultFS points to a local HDFS built on the instance storage which is temporary, with persistent data being stored in S3. Example: s3://wandisco.

fs.fusion.server

The hostname and request port of the Fusion server. Comma-separated list of $HOSTNAME:$PORT for multiple Fusion servers in the Zone.

fs.fusion.impl

The FileSystem implementation to be used for the fusion:/// URI scheme. See the Hadoop File System Configuration section for more detailed information.

fs.AbstractFileSystem.fusion.impl

The abstract filesystem implementation to be used. See General settings (core-site.xml) for further explanation.

5.3. Microsoft Azure Installation

This section covers how WANdisco Fusion can be used with Microsoft’s Cloud platform.

WANdisco Fusion supports the following storage types:
Azure Data Lake Storage Gen1 (ADLS Gen1)
Azure Data Lake Storage Gen2 (ADLS Gen2)
Azure Blob Storage (WASB)

5.3.1. Limitations and considerations

HDI 4.0 does not support ADLS Gen1 or WASB

WANdisco Fusion 2.14.1 and HDI 4.0 will only work for ADLS Gen2.

HDI 4.0 should not be used with ADLS Gen1 - this is a Microsoft limitation. For WASB, use Fusion 2.14.1 and HDI 3.6.

Secure transfers

If secure transfers are enabled in Azure then you must change the WASB scheme to WASBS in the following properties in /etc/hadoop/conf/core-site.xml:

Before:

fs.fusion.underlyingFs
wasb://container@storage.blob.core.windows.net

fs.defaultFS
wasb://container@storage.blob.core.windows.net

After:

fs.fusion.underlyingFs
wasbs://container@storage.blob.core.windows.net

fs.defaultFS
wasbs://container@storage.blob.core.windows.net

See the core-site.xml properties section for more information on these properties.

Owner access required

WANdisco Fusion requires that the Active Directory app is configured to run with owner-level service principals. To confirm you are the owner, open the Data Lake store account in the Azure portal and click on Access Control (IAM). Your service principal should show up on the list with the role set as "Owner".

As the Fusion Server now has owner-level service principals, any operation delegated to it by the client will be permitted. The Fusion Client performs some basic client-side access checks to ensure the client is permitted to do the operation before sending requests to the server. No checks can be done on the server side, however, as the Server uses its own service principals.

Additional core-site properties for ADLS-ADLS replication

If using ADLS to ADLS replication, and you want Consistency Checks to include the user/group and permission of files, then two core-site properties need to be added and set to true.

/etc/wandisco/fusion/server/core-site.xml
fusion.check.ownerAndGroup=true
fusion.check.permission=true
/etc/wandisco/fusion/ihc/server/<hdi-version>/core-site.xml
fusion.check.ownerAndGroup=true
fusion.check.permission=true

See the Consistency Check and Bypass settings (core-site) section for more information on these properties.

Service principal user and group

For ADLS, information about the service principal is needed. The directory this information is obtained from is set by fusion.adl.ugi.check.path.
See the Azure Configuration section for more information on this property.

5.3.2. Installation procedure in Azure

Installation of WANdisco Fusion in Azure follows a very similar process to the standard WANdisco Fusion installation.

As part of any Fusion install, please ensure that all pre-requisites are met prior to running the installer. See Pre-requisites Checklist and Cloud Deployment Guide for details.

CLI

The first part of the installation is CLI based. Follow the steps in the Initial installation via the terminal section to do this.

You will need to use a HDI specific installer, for example, fusion-ui-server-hdi_<OS-version>_installer.sh.

Note the differences below when you have reached the specify the appropriate platform option:

Please specify the appropriate platform from the list below:

[0] hdi-3.6 Data Lake Store
[1] hdi-3.6 Data Lake Store Gen 2
[2] hdi-3.6
[3] hdi-4.0 Data Lake Store
[4] hdi-4.0 Data Lake Store Gen 2
[5] hdi-4.0

Which WANdisco Fusion platform do you wish to use?

Select the appropriate HDI version and the appropriate storage platform:

WASB

hdi-<version>

ADLS Gen1

hdi-<version> Data Lake Store

ADLS Gen2

hdi-<version> Data Lake Store Gen 2

UI

The next section of the installer is browser based. For this, follow the steps in the Browser-based configuration of WANdisco Fusion section. All steps are the same for Azure installation except for some additional fields which need to be completed on page 5 of the installer. The options will depend on the storage type chosen. See below for your relevant set up.

Azure storage (WASB)
Azure install - WASB
Figure 45. Azure install - WASB
Primary (KEY1) Access Key

When you create a storage account, Azure generates two 512-bit storage access keys, which are used for authentication when the storage account is accessed. By providing two storage access keys, Azure enables you to regenerate the keys with no interruption to your storage service or access to that service. The Primary Access Key is now referred to as Key1 in Microsoft’s documentation. You can get the KEY from the Microsoft Azure storage account.
This can be changed on the UI Settings page after installation.

WASB storage URI

This needs to be in the format wasb[s]://<containername>@<accountname>.blob.core.windows.net

Validation then checks that:

  • the URI is in the correct format

  • it is possible for WANdisco Fusion to read from the Blob store

  • it is possible for WANdisco Fusion to write data to the Blob store

ADLS Gen1
Azure install - ADLSg1
Figure 46. Azure install - ADLS Gen1
ADL store URI

Path of the ADLS account to synchronize with. This must be in the format adl://<path to storage>.

Mount Point

The file path in the ADLS file system for Fusion to use as root. This will need to match your HDI cluster if you wish to use HDI to replicate your data out of ADLS.
For example / or /cluster/example.

The mount point must be correct at installation if matching a HDI cluster.
You must enter the correct mount point here, changing it after installation may not be possible. The mount point needs to match between your Fusion Server(s) and HDI cluster.
Client ID

The full client ID of the Active Directory credential you wish to use with Fusion.

Credential

The authentication key of the Active Directory credential you wish to use with Fusion.

Refresh Token URL

This needs to be in the format https://login.microsoftonline.com/<id>/oauth2/token.
Note: If you are intending to enable SSL on the Fusion core/IHC, then the Root CA of the above URL (login.microsoftonline.com) must be included inside of the Keystore that the Fusion Server will be using.

Service Principal Name

The service principal name (SPN) to be used to authenticate with the Azure Data Lake Store (ADLS). This must be set as an owner of the ADLS.

Validation then checks:

  • ADLS URI is in a valid format.

  • Refresh Token URL is in a valid format.

  • Whether the ADLS is accessible.

  • Whether Fusion can read from the store.

  • Whether Fusion can write data to the store.

ADLS Gen2
Azure install - ADLSg2
Figure 47. Azure install - ADLS Gen2
Container

The blob container that should be used.

Account name

The name of the ADLS Gen2 storage account.

KEY1 Access Key

When you create a storage account, Azure generates two 512-bit storage access keys, which are used for authentication when the storage account is accessed. By providing two storage access keys, Azure enables you to regenerate the keys with no interruption to your storage service or access to that service. The Primary Access Key is now referred to as Key1 in Microsoft’s documentation. You can get the KEY from the Microsoft Azure storage account.
This can be changed on the UI Settings page after installation.

Use SSL

Enable if the connection to the storage is to be SSL encrypted.

5.4. Google Cloud Installation

WANdisco’s Fusion can be installed on Google Cloud, enabling you to replicate on-premises data over to the Google Cloud Platform.

As well as using WANdisco Fusion with the Cloud Platform, you can also integrate it with Cloud Dataproc, Google’s cloud service for running Hadoop clusters.

This section takes you through how to install WANdisco Fusion on the Google Cloud Platform and then how to integrate it to Dataproc.

5.4.1. Connect WANdisco Fusion with your Google Cloud Storage

To use WANdisco Fusion with the Google Cloud Platform you first need to set up a VM. This guide assumes that you are already using Google Cloud and so have Bucket storage already set up. For more information see Google’s documentation.

  1. Log in to the Google Cloud Platform. Under VM instances in the Compute Engine section, click Create instance.

    WANdisco Fusion
    Figure 48. VM Instances
  2. Set up suitable specifications for the VM.

    WANdisco Fusion
    Figure 49. Create an instance
    Machine type

    2vCPUs recommended for evaluation.

    Boot disk

    Click on the Change button and select Centos 7.

    Increase Boot disk size

    Ensure that the boot disk size is sufficient.

    WANdisco Fusion
    Figure 50. Boot disk info
    Identity and API access

    Select 'Allow full access to all Cloud APIs'

    Firewall

    Enable publicly available HTTP and HTTPS.

  3. Expand the Management, disks, networking, SSH keys section and complete the following sections:

    Management

    On the Management tab, scroll to the Metadata section:

    • Specify the key, in this case startup-script-url

    • Add the value. This is the URL to the install script, the version number will need updating accordingly.
      For 2.14.0 - https://storage.googleapis.com/wandisco-public-files/2.14.0/installScript.sh

      WANdisco Fusion
      Figure 51. Metadata - install script
      Networking

      Select your Google Network VPC from the dropdown list and press Done.

      WANdisco Fusion
      Figure 52. Network
  4. Click Create to create your VM instance.

  5. When the instance is complete a green tick will appear on the VM Instances list.

    WANdisco Fusion
    Figure 53. Network
  6. Take note of the IP address of the instance. You will need this in the next step to access the UI for the WANdisco Fusion installer. Your network configuration will determine if you need to use the internal or external IP.

5.4.2. Installation of WANdisco Fusion for use with Google Cloud

Now you have created a VM on Google Cloud Platform you need to install WANdisco Fusion at your on-premises location.

To do this, open a browser and go to http://your-server-IP:8083/ to access the UI for your installer - this the IP noted in the previous step. Now follow the steps in the On premises installation section. When installing for use with Google Cloud, different information is required at step 5 in the installer compared to the standard installation. These additional fields are highlighted below:

Google
Figure 54. Step 5 - Zone information (Google Cloud deployment)
Google Bucket Name

The name of the Google storage bucket that will be replicated. This field will auto-fill with the name given in the Google Cloud Platform set up.

Google Project ID

The Google Project associated with the deployment. This field should auto-fill.

The following validation is completed against the settings:

Valid Bucket Name

Checks the provided bucket matches with an actual bucket on the platform.

Bucket Readable

Checks the bucket can be read by WANdisco Fusion.

Bucket writable

Checks the provided bucket can be written to by WANdisco Fusion.

Also, as there are no clients to install for this Cloud Platform you can skip step 7 of the installer.

For more information about completing cloud or hybrid deployments, see the Cloud Deployment Guide.

5.4.3. Cloud Dataproc

This section goes through how to integrate your on Cloud WANdisco Fusion node with Dataproc. Familiarity with how to use Google Cloud and Dataproc is assumed - visit the Google documentation on Dataproc if you need more information.

Prerequisites

Before you can use WANdisco Fusion with Dataproc you must first have:

Setting up Dataproc
  1. Download installClient.sh and complete the following properties at the top of the script. The rest of the script should not be altered.
    Note: this script is for 2.12.4.

    • GS_BUCKET_NAME= the name of the bucket your Fusion VM is synced with.

    • FUSION_NODE_IP= the IP address of your Fusion node.

      This completed script needs to be accessible to your Dataproc cluster.

  2. Now upload the script to your bucket. You can do this on the Google Cloud Platform UI in the Storage section.

    Dataproc
    Figure 55. Upload script to bucket - Dataproc

    Take note of the internal link of this script e.g. gs://yourbucket/installClient.sh, you will need this later.

  3. You now need to create a Dataproc cluster. To do this click on Dataproc in the Google Cloud Platform UI menu and then Create cluster.

    Dataproc
    Figure 56. Create cluster - Dataproc
  4. Choose a name for your cluster and fill out all the fields required for your set up.

    Dataproc
    Figure 57. Create cluster info - Dataproc
  5. Click Preemptible workers, bucket, network, version, initialization & access options to reveal more options.

    Dataproc
    Figure 58. Create cluster info - Dataproc

    Fill in the following information:

    Network

    Select your network from the drop down

    Image Version

    Select the version you are using - this will either be 1.0 or 1.1.

    Initialization actions

    The internal link to the completed installClient.sh in your bucket.

    Then click Create.

  6. Congratulations! You have integrated your WANdisco Fusion node with Dataproc.

5.4.4. Networking Guide for WANdisco Fusion Google Cloud

Setting up suitable network connectivity between your WANdisco Fusion zones using your Google Cloud private network system can be difficult to understand and implement if you’re not very familiar with the networking part of the Google Cloud platform. This section of the appendix will give you all the information you need to make the best choices and right configuration for setting up network connectivity between your on-premises and Google Cloud environments.

WANdisco Fusion makes the replication of your on-premises data to Google cloud simple and efficient. It relies on network connectivity between the two environments, and because a typical on-premises data store will reside behind your firewall, you will need to plan the right approach for that connection.

The following sections give information on the requirements that your solution will need to meet, along with options for establishing communication between the environments. You can choose among those options, and perform simple testing to ensure that the configured solution meets all your needs for data replication with WANdisco Fusion.

See the Cloud Deployment Guide for more information about setting up connections between cloud and on-premises WANdisco Fusion servers.

5.5. S3 Installation

WANdisco Fusion can be installed for use with S3, and S3 compatible platforms, enabling you to replicate data to cloud object storage.

5.5.1. Prerequisites for S3

See the Prerequisites Checklist for information about what is required for a WANdisco Fusion set up.

There is also more information on, for example, networking in the Cloud Deployment Guide.

If you wish to use Access and Secret keys, you may need to set up an AWS profile on the command line, see Setting up AWS profiles for guidance.

You can also read Amazon’s documentation about Getting your Access Key ID and Secret Access Key.

5.5.2. Installation of Fusion for use with S3 compatible storage

Installing WANdisco Fusion for use with S3 compatible storage follows a very similar process to the standard installation.

CLI

The first part of the installation is CLI based. Follow the steps in the Initial installation via the terminal section to do this.

You will need to use a S3 specific installer, for example, fusion-ui-server-s3_<OS-version>_installer.<fusion-version>.sh.

Note the differences below when you have reached the specify the appropriate platform option:

Please specify the appropriate platform from the list below:

[0] all s3 versions

Which WANdisco Fusion platform do you wish to use?

Select the only option available in this case to continue.

UI

The next section of the installer is browser based. For this, follow the steps in the Browser-based configuration of WANdisco Fusion section. All steps are the same for S3 installation except for some additional fields which need to be completed on page 5 of the installer.

Amazon install - S3
Figure 59. Amazon install - S3
Bucket Name

The name of the S3 Bucket that will connect to Fusion.

Bucket Region Endpoint

The endpoint Fusion should use to make requests to the S3 bucket. Set a local one to your region to reduce latency.

Bucket Region

The region in which your bucket was created.

Use Path based access

Enables the path style URL format for S3 buckets, see Amazon’s documentation examples for guidance if you are unsure.

Force V2 Auth Signing for S3

Older S3-compatible platforms may require Version 2 Authentication, and this can be enabled by checking the box. See Amazon’s documentation for further information.

Segment Size

The object size at which to split large objects into segments. Must be between 5MB and 5120MB (default is 128MB).

Buffer Directory

The directory used by the IHC Server for writing files locally before being uploaded to S3. Default is /opt/wandisco/fusion-ui-server/tmp.
This is required as data is not directly streamed into S3. The data stored in the buffer directory will be automatically cleaned up after ingest into S3 is complete.

Use access key and secret key

If you are using access key and secret key then additional fields are required. See the Use access key and secret key section below for more information on this. These settings can also be updated on the Settings tab in the UI after installation. This section is targeted at customers using Amazon S3 and so modifications may be needed depending on your specific type of S3.

Click Validate after entering the details so that the checks are performed to ensure that Fusion can correctly interact with the S3 Bucket. This may need to be performed after entering Access/Secret key details (if applicable).

Use access key and secret key

Access and secret keys are used by many S3 compatible storages. Tick this checkbox if the system user does not have authority to access the S3 storage system by, for example, insufficient IAM role permissions.

This option will reveal additional entry fields:

Access Key Id

This is your Access ID token for the service to use to communicate with the account. Validation tests that there is a provided value, along with a valid secret key.

Secret Key

The secret key token for the service to use to communicate with the account. It is used in conjunction with your Access Key ID to sign programmatic requests that are sent to your S3 storage.

Credentials File Path (Optional)

File containing the access and secret key, must be the absolute path to the file. Default is $HOME/.aws/Credentials, $HOME being the Fusion system user’s home directory (the system user for Fusion was set during the CLI phase).
Validation checks that the credentials file is accessible.

Profile

Select the profile associated with the Access/Secret key for the bucket. Credential profiles allow you to share multiple sets of security credentials between different tools like the AWS SDK for Java and the AWS CLI.

Click Validate to once all details are entered.

Configuration for S3

Once installation is complete, please see the S3 section in the Reference Guide for more information on additional properties that may need consideration.

You can also configure some properties on the Settings tab in the UI. These include:

S3 - Throttle Settings
Figure 60. S3 Throttle Settings
Retry before throttling

This toggles the throttling of the Amazon HTTP retries on and off.

Maximum number of retries

The maximum number of consecutive retries before throttling kicks in. The default is zero.

You can also adjust the Segment Size and Buffer Directory, which were defined during the UI phase of the installation.

5.6. Swift Installation

5.6.1. Installing into Openstack Swift storage

This section runs through the installation of WANdisco Fusion into an Openstack environment using Swift storage. We use Bluemix, the IBM cloud managed Swift solution, as an example but other implementations are available.
Currently this deployment is limited to an active-passive configuration that would be used to ingest data from your on-premises cluster to your Swift storage.

5.6.2. Pre-requisites

Before you begin an installation you need to have a Bluemix (or equivalent) account with container(s) set up. This guide runs through installing WANdisco Fusion and using it with Bluemix, but not how to set up Bluemix.

Make sure that you have the following directories created and suitably permissioned. Examples:

Important!
For installations to Swift storage, we currently only support Keystone 3.0.

5.6.3. Overview

The installation process runs through the following steps:

  1. On-premises installation - installing a WANdisco Fusion node on your cluster

  2. Setting up replication - Configure the nodes to ingest data from the on-premises cluster to the OpenStack Swift storage.

  3. Silent Installation - Notes on automating the installation process.

  4. Parallel Repairs - Running initial repairs in parallel.

5.6.4. Installation of WANdisco Fusion for use with Swift

To install WANdisco Fusion for Swift follow the steps in the On-premises installation guide. There a few differences when using Swift compared to the standard installation and these are highlighted below.

  • Make sure that you use a Swift specific installer, for example fusion-ui-server-swt_rpm_installer.sh.

Step 5 in the installer requires additional information compared to the standard installation. These additional fields are highlighted below:

WANdisco Fusion Deployment
Figure 61. Swift Install Information

Some of the required information can be gathered from the Bluemix UI (or other Swift Implementation), in the Service Credentials section:

WANdisco Fusion Deployment
Figure 62. Bluemix Credentials
User ID

The unique ID for the Bluemix/Swift user.

Password

The password for the Bluemix/Swift user.

Swift password changes
During installation, the Bluemix/Swift password is encrypted for use with WANdisco Fusion. This process doesn’t require any further interaction except for the case where the Swift password is changed. If you change your Swift password you need to do the following:

  1. Open a terminal to the WANdisco Fusion node and navigate to /opt/wandisco/fusion/tools/bin.

  2. Run the following script:

    ./encrypt-password.sh
      Please enter the password to be encrypted

    Enter your Bluemix/Swift password and press return:

    > password
    eCefUDtgyYczh3wtX2DgKAvXOpWAQr5clfhXSm7lSMZOwLfhG9YdDflfkYIBb7psDg3SlHhY99QsHlmr+OBvNyzawROKTd/nbV5g+EdHtx/J3Ulyq3FPNs2xrulsbpvBb2gcRCeEt+A/4O9K3zb3LzBkiLeM17c4C7fcwcPAF0+6Aaoay3hug/P40tyIvfnVUkJryClkENRxgL6La8UooxaywaSTaac6g9TP9I8yH7vJLOeBv4UBpkm6/LdiwrCgKQ6mlwoXVU4WtxLgs4UKSgoNGnx5t8RbVwlrMLIHf/1MFbkOmsCdij0eLAN8qGRlLuo4B4Ehr0mIoFu3DWKuDw==
    [ec2-user@ip-172-29-0-158 server]$
  3. Place the re-encrypted password in *core-site.xml* and *application.properties*.

Auth URL

The URL required for authenticating against Swift.

Swift Container Name

The name of the Swift storage container that Fusion will be connecting to.

Project Id

The Bluemix project ID.

Domain Name

The Swift Domain Name.

Segment Container

The name of the Segment container. The Segment container is used where large files break Swift’s 5GB limit for object size. Objects that exceed 5GB are broken into segments and get stored in here.

Region

The Swift Object Storage Region. Not to be confused with the Bluemix region.

Once you have entered this information click Validate.
The following Swift properties are validated:

Authorization URL reachable

Can you reach the keystone v3.0 authorization URL

Account valid

The installer checks that the Swift account details are valid. If the validation fails, you should recheck your Swift account credentials.

Container valid

The installer confirms that a container with the provided details exists. If the validation fails, check that you have provided the right container name.

Container readable

The container is checked to confirm that it can be read. If the validation fails, check the permissions on the container.

Container writable

The container is checked to confirm that the container can be written to. If the validation fails, check the permissions on the container.

The installer checks that the Swift account details are valid for accessing the segment container. If the validation fails, you should recheck your Swift account credentials.

Segment Container valid

The installer confirms that a segment container with the provided details exists. If the validation fails, check that you have provided the right segment container name.

Segment Container readable

The container is checked to confirm that it can be read. If the validation fails, check the permissions on the segment container.

Segment Container writable

The container is checked to confirm that the container can be written to. If the validation fails, check the permissions on the segment container.

Segment Account writable

The Account is checked to confirm that it can be written to. If the validation fails, check the permissions on the segment account.

In step 7 of the UI installer there are no clients to install so you can skip this step. The step is reserved for deployments where HDFS clients need to be installed.

For more information about the network configuration required for cloud or hybrid deployments, see the Cloud Deployment Guide.

5.6.5. Swift Silent Installation

You can complete a Swift installation using the Silent Installation procedure, putting the necessary configuration in the swift_silent_installer.properties and swift_silent_installer_env.sh as described in the section that covers Silent Installation.

Swift-specific settings

The following environment variables required for Swift deployments.

###############################
# Swift Configuration
###############################

#Swift installation mode
# REQUIRED for Swift Installations. Defaults to false
swift.installation.mode=true

#The Swift container name to use
# REQUIRED for Swift installations.
swift.containerName=

#The Swift userID to use to use
# REQUIRED for Swift installations.
swift.userID=

#The Swift password to use
# REQUIRED for Swift installations.
swift.password=

#Use HTTPS with the swift auth url
# REQUIRED for Swift installations.
swift.useHttps=false

#The Swift fully qualified domain name to use for authenticating access to the storage
# REQUIRED for Swift installations.
swift.auth.url=

# The Swift domain name to use
# REQUIRED, for Swift installations.
swift.domainName=

# The Swift project id to use
# REQUIRED, for Swift installations.
swift.projectId=

# The Swift file segment container to use
# REQUIRED, for Swift installations.
swift.segment.container=

# The Swift region to use
# OPTIONAL for Swift installations.
# swift.region

# The Swift buffer directory to use
# OPTIONAL for Swift installations, defaults to /tmp.
# swift.buffer.dir=

# The Swift  to use
# OPTIONAL for Swift installations, defaults to 5368709120 bytes, max 5368709120.
# swift.segment.size=

###############################
# Management Endpoint section
###############################

#The type of Management Endpoint.
management.endpoint.type=UNMANAGED_SWIFT
  • FUSIONUI_INTERNALLY_MANAGED_USERNAME

  • FUSIONUI_INTERNALLY_MANAGED_PASSWORD

  • FUSIONUI_FUSION_BACKEND_CHOICE

  • FUSIONUI_USER

  • FUSIONUI_GROUP

  • SILENT_PROPERTIES_PATH

Example Installation

As an example (as root), running on the installer moved to /tmp.

# If necessary download the latest installer and make the script executable
 chmod +x /tmp/installer.sh
# You can reference an original path to the license directly in the silent properties but note the requirement for being in a location that is (or can be made) readable for the $FUSIONUI_USER
# The following is partly for convenience in the rest of the script
cp /path/to/valid/license.key /tmp/license.key

# Create a file to encapsulate the required environmental variables:
cat <<EOF> /tmp/swift_silent_installer_env.sh
export FUSIONUI_MANAGER_TYPE=UNMANAGED_SWIFT
export FUSIONUI_INTERNALLY_MANAGED_USERNAME=admin
export FUSIONUI_FUSION_BACKEND_CHOICE=
export FUSIONUI_USER=hdfs
export FUSIONUI_GROUP=hdfs
export SILENT_PROPERTIES_PATH=/tmp/swift_silent.properties
export FUSIONUI_INTERNALLY_MANAGED_PASSWORD=admin
EOF

 # Create a silent installer properties file - this must be in a location that is (or can be made) readable for the $FUSIONUI_USER:
cat <<EOF > /tmp/swift_silent_installer_env.sh
existing.zone.domain=
existing.zone.port=
license.file.path=/tmp/license.key
server.java.heap.max=4
ihc.server.java.heap.max=4
fusion.domain=my.s3bucket.fusion.host.name
fusion.server.dcone.port=6444
fusion.server.zone.name=twilight
swift.installation.mode=true
swift.container.name=container-name
induction.skip=false
induction.remote.node=my.other.fusion.host.name
induction.remote.port=8082
EOF

# If necessary, (when $FUSIONUI_GROUP is not the same as $FUSIONUI_USER and the group is not already created) create the $FUSIONUI_GROUP (the group that our various servers will be running as):
[[ "$FUSIONUI_GROUP" = "$FUSIONUI_USER" ]] || groupadd hadoop

#If necessary, create the $FUSIONUI_USER (the user that our various servers will be running as):
useradd hdfs

if [[ "$FUSIONUI_GROUP" = "$FUSIONUI_USER" ]]; then
  useradd $FUSIONUI_USER
else
  useradd -g $FUSIONUI_GROUP $FUSIONUI_USER
fi

# silent properties and the license key *must* be accessible to the created user as the silent installer is run by that user
chown hdfs:hdfs $FUSIONUI_USER:$FUSIONUI_GROUP /tmp/s3_silent.properties /tmp/license.key

# Give s3_env.sh executable permissions and run the script to populate the environment
. /tmp/s3_env.sh

# If you want to make any final checks of the environment variables, the following command can help - sorted to make it easier to find variables!
env | sort

# Run installer:
/tmp/installer.sh

5.6.6. How Swift handles large files

Swift containers can appear to have a file-size discrepancy, looking smaller than the sum of their stored files. The reason for this apparent discrepancy is given below.

Files that are smaller than the segment size of a container are, predictably, stored directly to the container - as would be expected. However, large files that are bigger than the container’s segment size are actually stored in a companion container that is used for segments and has the same name as the parent container with the suffix "_segments". Segmented files appear in the main container, although these are empty manifest objects that symlink to the segments that correspond to the file in question.

So, for measuring the actual volume of data stored in a Swift container, you must also take into account the size of the corresponding segment’s container.

Impact on Fusion replication

WANdisco Fusion uses the same rules when replicating to Swift, and so provides configuration parameters for the ContainerName, SegmentContainerName and Segment Size for files uploaded via replication. Clearly, decreasing segmentSize for a container could increase the apparent storage size discrepancy, as more file content is actually stored in the segment container.

5.7. Cloud Deployment Guide

The following section expands upon the various Cloud installation guides, providing information that will help with general issues.

5.7.1. Networking between on-premises and Cloud

Setting up suitable network connectivity between your WANdisco Fusion zones using your Cloud private network system can be difficult to understand and implement if you’re not very familiar with the networking part of the Cloud platform. This section will give you all the information you need to make the best choices and right configuration for setting up network connectivity between your on-premises and Cloud environments.

WANdisco Fusion makes the replication of your on-premises data to cloud simple and efficient. It relies on network connectivity between the two environments, and because a typical on-premises data store will reside behind your firewall, you will need to plan the right approach for connecting it to a cloud resource.

The following sections give information on the requirements that your solution will need to meet, along with options for establishing communication between the environments. You can choose among those options, and perform simple testing to ensure that the configured solution meets all your needs for data replication with WANdisco Fusion.

Networking in WANdisco Fusion

WANdisco Fusion is a distributed system, allowing multiple, separate storage systems to exchange data to replicate content. It includes a collection of services that communicate over the network, requiring the ability to establish and use TCP connections between one another.

WANdisco Fusion for Cloud environments includes components that are created and launched in a Virtual Private Cloud (VPC). Your on-premises WANdisco Fusion components will need to establish connections with these VPC-resident services, and will also need to accept incoming connections from them.

You have many choices for how to establish connectivity between your on-premises environment and the Cloud-based Fusion node.

Regardless of your choice, you will need to ensure that your network connectivity meets the requirements defined in this document. For information from VPC connectivity options, please refer to documentation for you specific cloud platform, for example Using VPC Networks - Google or Amazon Virtual Private Cloud Connectivity Options.

Unidirectional Networking

WANdisco Fusion supports a feature that permits a switch in the direction of networking between the WANdisco Fusion server and remote IHC servers. By default, network connections are created outbound to any remote IHC servers during data transfer. To overcome difficulties in getting data back through your organization’s firewalls, it is possible to have Fusion wait for and re-use inbound connections.

Only turn on Inbound connection if you are sure that you need the open your network to traffic from the IHC servers on remove nodes.

WANdisco Fusion tree
Figure 63. Unidirectional networking diagram

5.7.2. Ports

The diagram below shows the Fusion services and the ports they expose, that are used to replicate content between an on-premises local file system and Google Cloud.

WANdisco Fusion tree
Figure 64. Cloud storage ports

Take careful note of the need for TCP connections to be established in both directions between the hosts on which the Fusion and IHC servers execute. You need to allow incoming and outgoing TCP connections on ports:

6444

Fusion port handles all co-ordination traffic that manages replication. It needs to be open between all WANdisco Fusion nodes. Nodes that are situated in zones that are external to the data center’s network will require unidirectional access through the firewall.

8023

Port used by WANdisco Fusion server to communicate with HCFS/HDFS clients. The port is generally only open to the local WANdisco Fusion server, however you must make sure that it is open to edge nodes.

8082

REST port is used by the WANdisco Fusion application for configuration and reporting, both internally and via REST API. The port needs to be open between all WANdisco Fusion nodes and any systems or scripts that interface with WANdisco Fusion through the REST API.

7000

7000 range, (exact port is determined at installation time based on what ports are available), used for data transfer between Fusion Server and IHC servers. Must be accessible from all WANdisco Fusion nodes in the replicated system.

8083

Used to access the WANdisco Fusion Administration UI by end users (requires authentication), also used for inter-UI communication. This port should be accessible from all Fusion servers in the replicated system as well as visible to any part of the network where administrators require UI access.

6. Operation

6.1. Initial configuration

Once WANdisco Fusion has been installed on all Clusters/Zones, you can log in to the Fusion UI for each node. The first time you log in the username and password are those set during installation (the Fusion Administrator Configuration section). After this the cluster manager details can be used to log in.

If this is a Hadoop cluster installation, ensure that the Cluster Manager details are entered first (the Complete your settings banner should then appear).

The Fusion UI will highlight tasks in the Complete your settings banner that need to be completed prior to setting up replication between Zones.

Dashboard - Configure Fusion
Figure 65. Dashboard - Configure WANdisco Fusion

Amber warnings may also be displayed that recommend enabling additional features (such as Kerberos authentication for the Fusion Core Server).

6.1.1. Steps required before Live replication

  1. Configure WANdisco Fusion by selecting the tasks or the specified section below on the Settings page.
    These tasks will need to be carried out on each Fusion in the Zone.

    For sections listed in this step, the Update option must be selected even if no changes have been made. The related task should disappear from the Complete your settings banner afterwards.

  2. Distribute Packages.
    Choose either option a or b depending on the URI Selection that was chosen for the Cluster/Zone.

    1. NameNode Proxy with HDFS

      1. Publish settings via the Fusion UI so that the local Cluster’s core-site and Fusion properties are updated.

        Settings - Publish Settings
        Figure 66. Publish Your Changes

        Ensure that this is performed on all Fusion nodes in the local Zone.
        If you get the warning Failed To Publish Your Settings, see the Troubleshooting section.

      2. Cluster Interaction

        • Choose one of the NameNode Proxy options within this section depending on platform.

    2. Fusion Client (all other URI Selection options)

      1. Cluster Interaction

        • Choose one of the Client Distribution options within this section depending on platform.

      2. Publish settings via the Fusion UI so that the local Cluster’s core-site and Fusion properties are updated.

        Settings - Publish Settings
        Figure 67. Publish Your Changes

        Ensure that this is performed on all Fusion nodes in the local Zone.
        If you get the warning Failed To Publish Your Settings, see the Troubleshooting section.

  3. Restart local Cluster services.
    A restart of local cluster services will now be required. This action is needed to activate the published configuration.

    If installing NameNode Proxy on Cloudera, this step will have already been completed. The only additional requirement is to restart the Cloudera Management Service as it will show a stale configuration from installing the NameNode Proxy CSD file.
  4. Restart Fusion services on all Fusion nodes in the Zone.

    • This is required for the Fusion Server, the IHC Server will need to be started for the first time. These operations can be performed within the Fusion UI, select each node in turn on the Nodes tab.

      Known issue in Fusion 2.14.1

      If the Fusion UI is displaying the Start IHC Server notification and the start/stop buttons are grayed out, check the status of the IHC Server manually in a terminal session on the Fusion node:

      service fusion-ihc-server-<distro-version> status

      If the status is not running, start the IHC Server manually using:

      service fusion-ihc-server-<distro-version> start

      Then restart the Fusion UI as this will enable the appropriate start/stop buttons.

      Additionally, if any of the IHC Servers are not displaying at all, restart all Fusion services manually on the terminal:

      service fusion-server restart
      service fusion-ihc-server-<distro-version> restart
      service fusion-ui-server restart
    • A restart of the Fusion UI service should be performed in a terminal session on the Fusion node.

      service fusion-ui-server restart

      If SSL has been enabled on the UI, please remember to adjust the URL to use the https prefix and the correct port (default is 8443).

  5. Enable live replication on the cluster.
    Choose either option a or b depending on the URI Selection that was chosen for the Cluster/Zone.

    1. NameNode Proxy with HDFS

      1. If NameNode Proxy High Availability (HA) is required, proceed to the following sections depending on platform:

      2. Read through the additional configuration section before proceeding to the next step.

      3. Once complete (or if NameNode Proxy HA is not required), proceed to enablement section depending on platform:

    2. Fusion Client (all other URI Selection options)
      Most of the standard Hadoop applications will have automatically been configured to work with Fusion during install, but some may require manual changes (including any third party applications).

Once all tasks have been completed, induction of the Fusion Zones will be required. See the Induction section for guidance.

After the ecosystem has been set up with at least two zones inducted together, replication rules can be created using the Fusion UI. Proceed to the Setting up Replication section for guidance on setting up rules.

6.1.2. Cluster Manager configuration

Configure the Cloudera or Ambari manager details, whichever is relevant to your set up.

Cluster Manager - CDH
Figure 68. Cloudera Manager Configuration
Cluster Manager - HDP
Figure 69. Ambari Configuration
Manager Host Name / IP

The FQDN for the server the manager is running on.

Port

The TCP port the manager is served from. The default is 7180 for Cloudera Manager and 8080 for Ambari.

Username

The username of the account that runs the manager. This account must have admin privileges on the Management endpoint.

Password

The password that corresponds with the above username.

  • Use SSL

    Tick the SSL checkbox if your Cluster Manager’s hostname uses HTTPS. You will be prompted to update the Port if you enable SSL, update if required.

    If using SSL the Ambari server certificate needs to be added to the Fusion UI SSL TrustStore (see UI TrustStore Options for which option was selected). Example below:

    Java keystore = /usr/java/jdk1.8.0_131/jre/lib/security/

Once you have entered the information, click Validate (see below for further elaboration). Once validated, select to Update.

The following aspects will be validated:

Cluster manager type

Validates connectivity with the cluster manager.

HDFS service state

Fusion validates that the HDFS service is running. If it is unable to confirm the HDFS state, a warning is given that will tell you to check the UI logs for possible errors. See the Logs section for more information.

HDFS service health

Fusion validates the overall health of the HDFS service. If the installer is unable to communicate with the HDFS service then you’re told to check the Fusion UI logs for any clues.
See the Logs section for more information.

HDFS service maintenance mode

Fusion looks to see if HDFS is currently in maintenance mode. Both Ambari and Cloudera support this feature. It is often when you need to make changes to your Hadoop configuration or hardware, as it will suppress alerts for a host, service, role or, if required, the entire cluster.
Fusion does not require maintenance mode to be off, this validation is simply to bring the state to your attention.

Fusion node as HDFS client

Validates that this Fusion node is a HDFS client.

6.1.3. Fusion Connection

The Fusion connection settings allows the enablement of SSL encryption between Fusion servers and IHC servers.

Fusion Server Settings - HTTP
Figure 70. Fusion Server Settings - HTTP
Fusion HTTP Policy Type

Sets the policy for communication with the Fusion Core Server API.

Only HTTP - WANdisco Fusion will not use SSL encryption on its API traffic.
Only HTTPS - WANdisco Fusion will only use SSL encryption for API traffic.
Use HTTP and HTTPS - WANdisco Fusion will allow both encrypted and un-encrypted traffic.

Fusion HTTP Server Port

The TCP port used for standard HTTP traffic. Validation checks whether the port is free and that it can be bound.

If configuring Fusion for the first time, the Update button can be selected if not enabling SSL at this time. If you are enabling SSL, follow the guidance in the Enable SSL for the WANdisco Fusion Server / IHC Server section.

6.1.4. Cluster Kerberos configuration

Enter the Kerberos details applicable to your environment.

Cluster Kerberos - disabled
Figure 71. Cluster Kerberos - Kerberos disabled
Kerberos disabled

When Kerberos is not enabled, this element will appear like the screenshot above.

Cluster Kerberos
Figure 72. Cluster Kerberos - Kerberos enabled
Kerberos enabled

When Kerberos is enabled, this element will appear like the screenshot above.

Configuration file path

The local file system path to the Kerberos krb5 configuration file. Normally installed in the /etc directory.

Keytab file path

The path to the Kerberos keytab file.

Handshake Token Directory

An optional entry. It defines what the root token directory should be for the Kerberos Token field. If left unset, by default, handshake tokens are created in the user’s working directories, e.g. /user/<username>/.

Click Validate. The Principal field will then become available as a drop-down list.

Principal

The unique identity to which Kerberos will assign authentication tickets.

Ensure the correct principal is selected. If the principal is changed, you will need to click Validate again.

Select Update once this is completed.

6.1.5. Fusion Kerberos configuration

Fusion provides an option to enable HTTP authentication on the Fusion Server API. The information in this section will provide guidance to enable this feature.

Fusion Kerberos - disabled
Figure 73. Fusion Kerberos - Kerberos disabled
Kerberos disabled

When Kerberos is not enabled, this element will appear like the screenshot above.

Fusion Kerberos
Figure 74. Fusion Kerberos - Kerberos enabled
Kerberos enabled

When Kerberos is enabled, this element will appear like the screenshot above.

Enable HTTP Authentication

Select to toggle whether HTTP authentication is enabled on the Fusion Server API.

The entries below will only appear if the checkbox is ticked.

Keytab file path

The path to the Kerberos keytab file that contains the HTTP principal.

Enable API Authorization

An optional entry. The Fusion server provides an authorization filter that is chained after the existing Kerberos authentication filter. This implementation makes use of the username supplied during authentication to check a list of users, stored in the application properties file and cluster core-site to decide if the action is authorized.

For details on how to configure users to be authorized to access the Fusion Server API, see the Configure users for Fusion API authorization section below.

Click Validate. Click Update once this is completed.

Configure users for Fusion API authorization

Once API Authorization is enabled in the Fusion Kerberos configuration, and Fusion settings have been published, users will need to be configured in two locations.

Fusion Servers
/etc/wandisco/fusion/server/application.properties
HDFS configuration
core-site.xml

The default properties and values specific to API authorization are listed below.

fusion.http.authorization.authorized.proxies=
fusion.http.authorization.authorized.read.writers=fusionUISystem,<fusion_superuser>
fusion.http.authorization.authorized.readers=fusionUISystem
fusion.http.authorization.authorized.proxies

Users that are allowed to become proxies on behalf of other users. HTTP calls (e.g. PUT) would include a value for the header "proxy.user.name". The proxied user’s permissions will then be checked against authorized reads and read-writers.

fusion.http.authorization.authorized.read.writers

Users that are allowed to make read OR write calls (i.e. any type of HTTP request). The user will be able to probe the Fusion API and make coordinated requests such as consistency checks, repairs and status calls.

fusion.http.authorization.authorized.readers

Users that are allowed to make read calls ONLY (write calls are PATCH, POST, PUT, DELETE).

  1. Add/remove any additional users to these properties in the HDFS configuration (core-site.xml) via the Cluster Manager. Ensure that the changes are saved.

  2. Edit the /etc/wandisco/fusion/server/application.properties to add/remove the same changes to the properties listed above. Ensure that this step is performed on all Fusion servers in the Zone.

  3. HDFS and dependent services will need to be restarted via the Cluster Manager to make these changes effective. It may be worth waiting until all additional changes have been published to the Cluster Manager before performing this.

  4. On all Fusion servers in the Zone, run the following commands to restart the Fusion services:

    service fusion-server restart
    service fusion-ihc-server-<distro-version> restart
    service fusion-ui-server restart

    This will enable the user changes that have been made.

6.1.6. Replication settings

The following configuration options are available in the Replication section of the Settings tab.

Replication settings
Figure 75. Replication Settings
Support ACL Replication

Fusion can support ACL replication. For Hadoop Compatible File Systems, this option will default to true. If ACL replication is required on a different filesystem, you can enable ACL replication support by ticking this checkbox.

To support ACL replication, the dfs.namenode.acls.enabled property must exist and be set to true in the HDFS configuration of the local cluster. See Apache Hadoop ACLs (Access Control Lists) for more details on this property and its usage.
Enable ACL replication

Toggle to enable ACL replication. If ACL replication is enabled, then changes from both local and remote zones are executed. If disabled, only locally originated ACL modification commands are executed.

  • This is a zone setting, so is enabled for the whole zone on which you enable the option.

  • Regardless of the state of this setting, a HDFS client loading FusionHdfs will submit agreements for File ACL changes in HDFS (if it is on a replicated path and not-excluded).

  • While a local zone will always execute a locally generated ACL change, it will only be executed on other zones, if the ACL Replication checkbox is ticked.

When toggling ACL replication on or off, a manual restart of the Fusion Core Server and the Fusion IHC Server will be required. It is recommended to wait until all configuration changes have been made first, especially if configuring Fusion for the first time.

service fusion-server restart
service fusion-ihc-server-<distro-version> restart

Replication Exchange Directory
Location of a directory in the replicated filesystem to which the Fusion Server will write information about replicated directories for clients to read. It should be a non-replicated location, readable by all users and writable by the Fusion user. If the directory doesn’t exist, it will be automatically created during the next Fusion server restart.

Use replication exchange directory

Checkbox (unticked by default).

Path to exchange directory

Entry field for the path to the exchange directory. The entry field only appears if you tick the checkbox.

Setting a replication exchange directory will trigger the following warning:

Changes to these settings require the manual restarting of the Fusion Core Server. You will also need to redistribute the client configurations.

If configuring Fusion for the first time, return to the Steps required before Live replication section to continue the guidance.

Otherwise, note the following steps required after configuring this feature.

  1. HDFS and dependent services will need to be restarted via the Cluster Manager to make these changes effective.

  2. On all Fusion servers in the Zone, run the following commands to restart the Fusion services:

    service fusion-server restart
    service fusion-ihc-server-<distro-version> restart

    The Fusion services should be restarted once HDFS and dependent services have finished restarting.

6.1.7. Cluster Interaction

This section will be dependent on the URI Selection chosen during installation.

Cluster Interaction
Figure 76. Cluster Interaction options
Fusion NameNode Proxy = NameNode Proxy with HDFS

See the following links for guidance depending on the platform:

Fusion Client Library = All other URI Selection options

See the following links for guidance depending on the platform:

NameNode Proxy - Hortonworks

After WANdisco Fusion has been successfully installed, you can proceed to distribute and then configure the NameNode Proxy (after having published Fusion configuration).

  1. NameNode Proxy packages need to be distributed before configuration is possible. Follow the on screen steps to download the Stack packages.

    CI: NameNode Proxy - Ambari
    Figure 77. CI: NameNode Proxy - Ambari
  2. Download the compressed stack onto the Ambari Server.

    Example
    Right-click the Stack package and copy the link address to use in the command below:

    wget <namenode_proxy_stack.tar.gz>
  3. Stop the Ambari Server.

    ambari-server stop
  4. Install the NameNode Proxy mpack.

    You must enter the absolute path to the stack package in single (') quotes.
    ambari-server install-mpack --mpack='/path/to/<namenode_proxy_stack.tar.gz>'
  5. Start the Ambari Server.

    ambari-server start
  6. Remove the compressed stack inside the services directory afterwards.

    rm -f /path/to/<namenode_proxy_stack.tar.gz>
  7. In the Ambari UI, select ServicesAdd ServiceNameNode Proxy.

    Press Next

  8. Choose the hosts where you want to install NameNode Proxy Master and press Next. If requiring NameNode Proxy HA, ensure that two hosts are selected for the NameNode Proxy Master.

    A common location for NameNode Proxy Master(s) is either the Fusion Servers or the Active/Standby NameNodes.
  9. In the Customize Services page, configure any of the NameNode Proxy properties if required. These can be left as default values.

    Once the properties have been configured, press Next to review your changes and deploy the installation.

  10. After the installation process is complete, return to the Steps required before Live replication section for guidance on publishing configuration and restarting services.

Configuring NameNode Proxy for High Availability (HA) in Ambari

There are two methods for implementing NameNode Proxy for HA in a HDFS environment:

  1. Create a new nameservice for the NameNode Proxies and point the applications to use this new nameservice.

  2. Repurpose the existing nameservice for the NameNode Proxies, and create a new nameservice to represent the NameNodes of the previous nameservice.

The choice will depend on whether there is a requirement to keep the current nameservice in place rather than replacing it with a new one. In either case, the fs.defaultFS property will eventually need to be set to the nameservice assigned to the NameNode Proxies.

The two methods are outlined below.

Create a new nameservice for the NameNode Proxies

For this example, the following variables have been defined:

Existing nameservice = nameservice01
New nameservice = nnproxies
NameNode Proxy 1 unique identifier = nnproxy1
NameNode Proxy 2 unique identifier = nnproxy2

  1. In Ambari, select HDFSConfigs → Search for Custom hdfs-site

  2. Select to Add Property and use the Bulk property add mode option to add the following properties (ensure to edit them to suit your environment).

    1. dfs.client.failover.proxy.provider.nnproxies=org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider

      • This defines two nameservices for this Hadoop cluster.

    2. dfs.ha.namenodes.nnproxies=nnproxy1,nnproxy2

    3. dfs.namenode.rpc-address.nnproxies.nnproxy1=<nnproxy1 address with port 8890>

    4. dfs.namenode.rpc-address.nnproxies.nnproxy2=<nnproxy2 address with port 8890>

    5. If using HDP 3.0 or greater, you can randomize the NameNode Proxy order with the following property (HDFS-6648):

      • dfs.client.failover.random.order.nnproxies = true

  3. Edit the following properties, or add them if they do not exist.

    1. dfs.nameservice.id=nameservice01

      • This defines the real NameNodes to be used when resolving the addresses that belongs to the NameNodes.

    2. dfs.internal.nameservices=nameservice01

      • This defines the NameNodes' nameservice to be used by internal services, such as Datanodes.

  4. Continue with either option a or b depending on how you wish to proceed.

    1. If wishing to enable NameNode Proxy immediately, proceed to the Enabling NameNode Proxy in Cloudera section for guidance.

    2. If not wanting to enable NameNode Proxy at this time, deploy all configurations and restart cluster services including NameNode Proxy. Please note that Live replication will not be possible until the steps in Enabling NameNode Proxy in Ambari are carried out.

Repurpose the existing nameservice for the NameNode Proxies, and create a new nameservice for the NameNodes.

For this example, the following variables have been defined:

Existing nameservice = nameservice01
New nameservice = nameservice02
NameNode Proxy 1 unique identifier = nnproxy1
NameNode Proxy 2 unique identifier = nnproxy2
NameNode 1 unique identifier = nn1
NameNode 2 unique identifier = nn2

It is highly recommended to perform these steps under maintenance conditions as it involves reconfiguring the underlying nameservice for the cluster.
  1. In Ambari, stop all cluster services except ZooKeeper.

  2. Remove the current nameservice using ZooKeeper.

    Known issues - ZOOKEEPER-2370 / ZOOKEEPER-729

    Please be aware of the known issues in ZooKeeper when carrying out the steps in this task. You may experience errors when trying to log into ZooKeeper or may fail to delete the required directory with an error such as: Authentication is not valid.

    Examples for workarounds to these issues are detailed in the Knowledgebase article - Workarounds for ZOOKEEPER-729/ZOOKEEPER-2370.

    1. Open a terminal session to a ZooKeeper server host and run zookeeper-client.

    2. Execute the following command to remove the configured nameservice:

      • rmr /hadoop-ha/nameservice01

  3. In Ambari, create the underlying nameservice configuration again except under a new name.

    1. Go to HDFSConfigsCustom hdfs-site.

    2. Select to Add Property and use the Bulk property add mode option to add the following properties (ensure to edit them to suit your environment).

      • dfs.client.failover.proxy.provider.nameservice02=<nameservice01’s FailoverProxyProvider>

      • dfs.ha.namenodes.nameservice02=nn1,nn2

      • dfs.nameservice.id=nameservice02

      • dfs.namenode.http-address.nameservice02.nn1=<nameservice01.namenode1’s http-address>:50070

      • dfs.namenode.http-address.nameservice02.nn2=<nameservice01.namenode2’s http-address>:50070

      • dfs.namenode.https-address.nameservice02.nn1=<nameservice01.namenode1’s https-address>:50470

      • dfs.namenode.https-address.nameservice02.nn2=<nameservice01.namenode2’s https-address>:50470

      • dfs.namenode.rpc-address.nameservice02.nn1=<nameservice01.namenode1’s rpc-address>:8020

      • dfs.namenode.rpc-address.nameservice02.nn2=<nameservice01.namenode2’s rpc-address>:8020

  4. Edit the following properties in the Custom hdfs-site so that the new nameservice is referenced correctly.

    1. dfs.internal.nameservices=nameservice02

      • This defines the NameNodes' nameservice to be used by internal services, such as Datanodes.

    2. dfs.nameservices=nameservice01,nameservice02

      • The nameservice01 will be repurposed in a later step for the NameNode Proxies.

    3. dfs.namenode.shared.edits.dir=qjournal://<qjournal hosts>/nameservice02

      • This property sets the addresses of the JournalNodes, which provide the shared edits storage. The property value should be of the form: qjournal://host1:port1;host2:port2;host3:port3/journalId.

  5. Ensure to Save the configuration before continuing.

  6. Select HDFSACTIONSRestart JournalNodes.

    • This will restart the JournalNodes with the configs deployed so that the initialize shared edits command can be run in a later step.

    • If possible, lower the rolling restart period to make these restarts faster.

  7. Refresh the configuration for the HDFS service.

    1. Select Hosts on the left panel in Ambari.

    2. Select the NameNode host that corresponds to the nn1 unique identifier.

    3. Under Action for HDFS Client / HDFS in the Components list, select to Refresh configs.

  8. Open a terminal session to the nn1 host and run the following commands.

    1. Run the initialize shared edits command (ensure it is run by the appropriate user).

      • hdfs namenode -initializeSharedEdits -force

    2. Run the format ZooKeeper command (ensure it is run by the appropriate user).

      • hdfs zkfc -formatZK -force

Now that the current NameNodes have been moved to nameservice02, the original nameservice (nameservice01) can be repurposed for the NameNode Proxies.

  1. In Ambari, select HDFSConfigs → Search for Custom hdfs-site.

  2. Either edit the following properties if they exist, or select to Add Property and use the Bulk property add mode option to add the following properties (ensure to edit them to suit your environment).

    1. dfs.nameservices = nameservice01,nameservice02

      • This defines two nameservices for this Hadoop cluster.

    2. dfs.ha.namenodes.nameservice01 = nnproxy1,nnproxy2

    3. dfs.namenode.rpc-address.nameservice01.nnproxy1 = <nnproxy1 address with port 8890>

    4. dfs.namenode.rpc-address.nameservice01.nnproxy2 = <nnproxy2 address with port 8890>

    5. dfs.nameservice.id = nameservice02

      • This defines the real NameNodes to be used when resolving the addresses that belongs to the NameNodes.

    6. dfs.internal.nameservices = nameservice02

      • This defines the NameNodes' nameservice to be used by internal services, such as Datanodes.

    7. If using HDP 3.0 or greater, you can randomize the NameNode Proxy order with the following property (HDFS-6648):

      • dfs.client.failover.random.order.nameservice01 = true

  3. In Ambari, adjust the fs.fusion.underlyingFs property so that it is set to the new NameNode nameservice.

    1. Select HDFSConfigs → Search for Custom core-site

    2. Change the value of fs.fusion.underlyingFs from nameservice01 to nameservice02.

      • Now that the underlying namenodes are represented by nameservice02, we needed to point the underlying FileSystem to it.

  4. It is now necessary to deploy all configurations and restart services including NameNode Proxy. Proceed to the Enabling NameNode Proxy in Ambari section for guidance.

Enabling NameNode Proxy in Ambari

If the existing nameservice was repurposed for the NameNode Proxies (documented here), then ignore the steps to change the fs.defaultFS value and proceed to the Restart order step.

  1. In the Ambari UI, select HDFSConfigs → and Filter for the property fs.defaultFS.

  2. Change the value of fs.defaultFS to point at either the NameNode Proxy host (default port is 8890) or NameNode Proxy HA nameservice depending on your environment.

    Example for NameNode Proxy on single host

    fs.defaultFS = <namenode_proxy_hostname>:<namenode_proxy_port>

    Example for NameNode Proxy HA

    fs.defaultFS = hdfs://<namenode_proxy_ha_nameservice>
  3. Once all of the configuration changes are completed, it is necessary to restart services in the correct order.

    Restart order
    1. Deploy new HDFS Client configuration on all NameNode Proxy and Fusion Servers.
      To perform this task, the Refresh configs option can be used on individual hosts in the Ambari UI.

    2. NameNode Proxy
      Restart the NameNode Proxy service in the Ambari UI.

    3. Fusion services
      Restart the Fusion Server and IHC Server in either the Fusion UI or via a terminal on the Fusion nodes.

      Example for terminal commands

      service fusion-server restart
      service fusion-ihc-server-<distro-version> restart
    4. HDFS and remaining cluster services.
      Perform the HDFS and remaining cluster service restarts in the Ambari UI.

NameNode Proxy - Cloudera

After WANdisco Fusion has been successfully installed, you can proceed to distribute and then configure the NameNode Proxy (after having published Fusion configuration).

  1. NameNode Proxy packages need to be distributed before configuration is possible. Follow the on screen steps to download the Parcel/CSD packages.

    CI: NameNode Proxy - Ambari
    Figure 78. CI: NameNode Proxy - Cloudera
  2. Download the CSD file onto the Cloudera Management Server.

    Example
    Right-click the CSD file and copy the link address to use in the command below:

    wget <NAMENODE_PROXY-CSD.jar>
  3. Move the CSD file into the Cloudera CSD directory.

    Example

    mv /path/to/NAMENODE_PROXY-CSD.jar /opt/cloudera/csd/
  4. Change the ownership of the CSD file so that it matches the system account that runs Cloudera Manager.

    chown cloudera-scm:cloudera-scm /opt/cloudera/csd/NAMENODE_PROXY-CSD.jar
  5. Download the compressed parcel onto the Cloudera Management Server.

    Example
    Right-click the NameNode Proxy Parcel and copy the link address to use in the command below:

    wget <namenode_proxy.parcel>
    Rename parcel if using RHEL 7/CentOS 7
    By default the parcels are el6. If using RHEL 7/CentOS 7, rename the parcels to el7. This will prevent Cloudera throwing an error around the expected parcel name.
  6. Download the compressed parcel sha onto the Cloudera Management Server.

    Example
    Right-click the parcel sha file and copy the link address to use in the command below:

    wget <namenode_proxy.parcel.sha>
  7. Move the parcel and sha files into the Cloudera parcel directory.

    Example

    mv /path/to/<namenode_proxy.parcel> /opt/cloudera/parcel-repo/
    mv /path/to/<namenode_proxy.parcel.sha> /opt/cloudera/parcel-repo/
  8. Change the ownership of the parcel and sha file so that they match the system account that runs Cloudera Manager.

    chown cloudera-scm:cloudera-scm /opt/cloudera/parcel-repo/NAMENODE_PROXY-*.parcel*
  9. Restart the Cloudera Manager.

    service cloudera-scm-server restart
  10. In the Cloudera UI, navigate to the Parcels screen.

  11. Locate the NAMENODE_PROXY and click on the Distribute button to install from the parcel.

  12. Once distributed, click on the Activate button to active from the parcel. Wait until activation is complete before continuing.

  13. Restart HDFS and cluster services.
    It is now necessary to deploy client configuration and restart cluster services via the Cloudera UI.

  14. Once the service restarts are complete, on the Cloudera Manager Home page, select to Add Service.

  15. Select the NameNode Proxy from the service list, and click Continue.

  16. On the Assign Roles step, select hosts for the NameNode Proxy Server to be installed. If requiring NameNode Proxy HA, ensure that two hosts are assigned this role.

    A common host(s) for NameNode Proxy Master(s) is either the Fusion Servers or the Active/Standby NameNodes.

    There is no requirement to install a Gateway on any host, so this part can be ignored. Click Continue after assignment.

  17. In the Review Changes step, the following properties will need to configured manually.

    Key Description Value Example

    zk-dt-secret-manager.zkConnectionString

    A comma separated list of Zookeeper hosts and ports to connect to.

    ZKHOST1:2181,ZKHOST2:2181,ZKHOST3:2181

    zk-dt-secret-manager.zkAuthType

    The authentication protocol used for messages sent to Zookeeper. May already be configured by default.

    none,sasl

    If the zk-dt-secret-manager.zkAuthType value is set to sasl, then the following two additional properties must be configured.

    Key Description Value Example

    zk-dt-secret-manager.kerberos.principal

    The Kerberos service principal used to connect to Zookeeper. Only necessary if zkauthType is sasl.

    hdfs@REALM.COM,hdfs/hostname@REALM.COM

    zk-dt-secret-manager.kerberos.keytab

    The absolute path for the Kerberos keytab with the credentials to connect to Zookeeper. Only necessary if zkauthType is sasl.

    /etc/security/keytabs/nnproxy.keytab

    Once the properties have been configured, press Continue to perform the test run. Click Continue once this has completed, and Finish on the Summary screen.

  18. After the installation process is complete, return to the Steps required before Live replication section for guidance on publishing configuration and restarting services.

Configuring NameNode Proxy for High Availability (HA) in Cloudera

There are two methods for implementing NameNode Proxy for HA in a HDFS environment:

  1. Create a new nameservice for the NameNode Proxies and point the applications to use this new nameservice.

  2. Repurpose the existing nameservice for the NameNode Proxies, and create a new nameservice to represent the NameNodes of the previous nameservice.

The choice will depend on whether there is a requirement to keep the current nameservice in place rather than replacing it with a new one. In either case, the fs.defaultFS property will eventually need to be set to the nameservice assigned to the NameNode Proxies.

The two methods are outlined below.

Create a new nameservice for the NameNode Proxies

For this example, the following variables have been defined:

Existing nameservice = nameservice01
New nameservice = nnproxies
NameNode Proxy 1 unique identifier = nnproxy1
NameNode Proxy 2 unique identifier = nnproxy2

  1. In Cloudera, select HDFSConfiguration → Search for HDFS Service Advanced Configuration Snippet (Safety Valve) for hdfs-site.xml

  2. Select to Add new properties (ensure to edit them to suit your environment).

    1. dfs.nameservices = nameservice01,nnproxies

      • This defines two nameservices for this Hadoop cluster.

      • It will overwrite Cloudera Manager’s auto-set dfs.nameservices.

    2. dfs.ha.namenodes.nnproxies = nnproxy1,nnproxy2

    3. dfs.namenode.rpc-address.nnproxies.nnproxy1 = <nnproxy1 address with port 8890>

    4. dfs.namenode.rpc-address.nnproxies.nnproxy2 = <nnproxy2 address with port 8890>

    5. dfs.nameservice.id = nameservice01

      • This defines the real NameNodes to be used when resolving the addresses that belongs to the NameNodes.

    6. dfs.internal.nameservices = nameservice01

      • This defines the NameNodes' nameservice to be used by internal services, such as Datanodes.

    7. If using CDH 6.0 or greater, you can randomize the NameNode Proxy order with the following property (HDFS-6648):

      • dfs.client.failover.random.order.nnproxies = true

  3. Repeat this process again except add the properties to HDFS Client Advanced Configuration Snippet (Safety Valve) for hdfs-site.xml instead.

  4. Continue with either option a or b depending on how you wish to proceed.

    1. If wishing to enable NameNode Proxy immediately, proceed to the Enabling NameNode Proxy in Cloudera section for guidance.

    2. If not wanting to enable NameNode Proxy at this time, deploy all configurations and restart cluster services including NameNode Proxy. Please note that Live replication will not be possible until the steps in Enabling NameNode Proxy in Cloudera are carried out.

Repurpose the existing nameservice for the NameNode Proxies, and create a new nameservice for the NameNodes.

For this example, the following variables have been defined:

Existing nameservice = nameservice01
New nameservice = nameservice02
NameNode Proxy 1 unique identifier = nnproxy1
NameNode Proxy 2 unique identifier = nnproxy2

It is highly recommended to perform these steps under maintenance conditions as it involves reconfiguring the underlying nameservice for the cluster.
  1. In Cloudera, stop all cluster services except ZooKeeper.

  2. Remove the current nameservice using Zookeeper.

    Known issues - ZOOKEEPER-2370 / ZOOKEEPER-729

    Please be aware of the known issues in ZooKeeper when carrying out the steps in this task. You may experience errors when trying to log into ZooKeeper or may fail to delete the required directory with an error such as: Authentication is not valid.

    Examples for workarounds to these issues are detailed in the Knowledgebase article - Workarounds for ZOOKEEPER-729/ZOOKEEPER-2370.

    1. Open a terminal session onto a ZooKeeper server host and run zookeeper-client.

    2. Execute the following command to remove the configured nameservice:

      • rmr /hadoop-ha/nameservice01

        The nameservice name can be found in the HDFSInstancesFederation and High Availability section.
  3. In the Cloudera Manager Admin Console, change the NameNode nameservice name to nameservice02.

    1. Go to the HDFSConfiguration.

    2. Type NameNode Nameservice in the Search field.

    3. For the NameNode Nameservice property, change the current nameservice to the nameservice02 name in all the fields (per node and Default Group). The name must be unique and can contain only alphanumeric characters.

    4. Type Quorum-based Storage Journal name in the Search field.

    5. For the Quorum-based Storage Journal property, change the current nameservice to the nameservice02 name in all the fields (per node and Default Group).

    6. Type Cluster-wide Advanced Configuration Snippet (Safety Valve) for core-site.xml in the Search field.

    7. Add a new property with name fs.defaultFS and value hdfs://nameservice01.

      • This is required because Cloudera Manager will automically change fs.defaultFS to nameservice02 when changing the NameNode Nameservice. Since we want to keep it as nameservice01, this will maintain the manual configuration.

    8. Type HDFS Service Advanced Configuration Snippet (Safety Valve) for hdfs-site.xml in the Search field.

    9. Add the following property (ensure to edit them to suit your environment).

      • dfs.client.failover.proxy.provider.nameservice02 = <nameservice02’s FailoverProxyProvider>

        • Search for dfs.client.failover.proxy.provider in the HDFS config to find the Service-Wide value for this property.

    10. Type HDFS Service Advanced Configuration Snippet (Safety Valve) for hdfs-site.xml in the Search field.

    11. Add the dfs.client.failover.proxy.provider.nameservice02 property again for this section.

    12. Click Save Changes to commit the configuration.

  4. Go to HDFSInstances.

    1. In the Federation and High Availability pane, select ActionsInitialize High Availability State in ZooKeeper.

      • This creates a znode to manage NameNode HA automatic failover for nameservice02.

  5. If Hive is installed on the cluster, perform the following steps. If Hive is not installed, proceed to the next numbered step.

    1. Go to the Hive service.

    2. Select ActionsUpdate Hive Metastore NameNodes.

  6. Go to HDFSInstances.

    1. Select the checkboxes next to the JournalNode role instances.

    2. Select Actions for SelectedStart.

    3. Click any NameNode role instance.

    4. Select ActionsInitialize Shared Edits Directory.

      • This initializes the shared edits directory for the new nameservice in Quorum Journal. This is required because, in step 3e, we specified the new storage location as nameservice02.

  7. Click the Cloudera Manager logo to return to the Home page.

  8. In the cluster actions list, select to Deploy Client Configuration.

  9. Start all cluster services (except ZooKeeper as it is already started).

Now that the current NameNodes have been moved to nameservice02, the original nameservice (nameservice01) can be repurposed for the NameNode Proxies.

  1. In Cloudera, select HDFSConfiguration → Search for HDFS Service Advanced Configuration Snippet (Safety Valve) for hdfs-site.xml

  2. Add the following properties (ensure to edit them to suit your environment).

    1. dfs.client.failover.proxy.provider.nameservice01 = <nameservice01’s FailoverProxyProvider>

      • Search for dfs.client.failover.proxy.provider in the HDFS config to find the Service-Wide value for this property.

    2. dfs.nameservices = nameservice01,nameservice02

      • This defines two nameservices for this Hadoop cluster.

      • It will overwrite Cloudera Manager’s auto-set dfs.nameservices.

    3. dfs.ha.namenodes.nameservice01 = nnproxy1,nnproxy2

    4. dfs.namenode.rpc-address.nameservice01.nnproxy1 = <nnproxy1 address with port 8890>

    5. dfs.namenode.rpc-address.nameservice01.nnproxy2 = <nnproxy2 address with port 8890>

    6. dfs.nameservice.id = nameservice02

      • This defines the real NameNodes to be used when resolving the addresses that belongs to the NameNodes.

    7. dfs.internal.nameservices = nameservice02

      • This defines the NameNodes' nameservice to be used by internal services, such as Datanodes.

    8. If using CDH 6.0 or greater, you can randomize the NameNode Proxy order with the following property (HDFS-6648):

      • dfs.client.failover.random.order.nameservice01 = true

  3. Repeat this process again except add the properties to HDFS Client Advanced Configuration Snippet (Safety Valve) for hdfs-site.xml instead.

  4. In Cloudera, adjust the fs.fusion.underlyingFs property so that it is set to the new NameNode nameservice.

    1. Select HDFSConfiguration → Search for Cluster-wide Advanced Configuration Snippet (Safety Valve) for core-site.xml

    2. Change the value of fs.fusion.underlyingFs from nameservice01 to nameservice02.

      • As the underlying NameNodes are now represented by nameservice02, this value change is required to point Fusion to the underlying filesystem.

  5. It is now necessary to deploy all configurations and restart services including NameNode Proxy. Proceed to the Enabling NameNode Proxy in Cloudera section for guidance.

Enabling NameNode Proxy in Cloudera

If the existing nameservice was repurposed for the NameNode Proxies (documented here), then ignore the steps to set the fs.defaultFS value and proceed to the Restart order step.

  1. In the Cloudera UI, select HDFSConfigurationSearch for the property fs.defaultFS.
    If the property is not found, add the property to the Cluster-wide Advanced Configuration Snippet (Safety Valve) for core-site.xml section.

  2. Set/change the value of fs.defaultFS to point at either the NameNode Proxy host (default port is 8890) or NameNode Proxy HA nameservice depending on your environment.

    Example for NameNode Proxy on single host

    fs.defaultFS = <namenode_proxy_hostname>:<namenode_proxy_port>

    Example for NameNode Proxy HA

    fs.defaultFS = hdfs://<namenode_proxy_ha_nameservice>
  3. Once all of the configuration changes are completed, it is necessary to restart services in the correct order.

    Restart order
    1. Deploy new HDFS Client configuration on all NameNode Proxy and Fusion Servers.
      To perform this task, the Deploy Client Configuration option can be used for the HDFS service in Cloudera. This will deploy the new HDFS client configuration for all HDFS Gateway nodes. Cloudera does not yet provide an option to refresh service config for individual hosts.

    2. NameNode Proxy
      Restart the NameNode Proxy service in the Cloudera UI.

    3. Fusion services
      Restart the Fusion Server and IHC Server in either the Fusion UI or via a terminal on the Fusion nodes.

      Example for terminal commands

      service fusion-server restart
      service fusion-ihc-server-<distro-version> restart
    4. HDFS and remaining cluster services.
      Perform the HDFS and remaining cluster service restarts in the Cloudera UI.

Additional configuration for NameNode Proxy

If installing Fusion for the first time, remember to return to the Steps required before Live replication guidance after reading through the sections below.

Configuring NameNode Proxy with Live Hive

If Live Hive is installed or will be installed on the cluster, then additional configuration is required to ensure it will be integrated correctly with the NameNode Proxy service.

  1. Navigate to the Hive Metastore configuration to add a property that will overwrite the core-site.xml (see option a or b depending on platform).

    1. Ambari
      HiveConfigsAdvancedCustom hivemetastore-site.

    2. Cloudera
      HiveConfigurationHive Metastore Server Advanced Configuration Snippet (Safety Valve) for core-site.xml.

  2. Add a property that will point the Hive Metastore(s) to the NameNodes (rather than the NameNode Proxies) by setting dfs.ha.namenodes.<nnproxies-nameservice> to reference the NameNodes' unique identifiers.

    This example uses the following variables:

    NameNode Proxy nameservice = nnproxies
    NameNode Proxy 1 unique identifier = nnproxy1
    NameNode Proxy 2 unique identifier = nnproxy2
    NameNode 1 unique identifier = nn1
    NameNode 2 unique identifier = nn2

    Before: dfs.ha.namenodes.nnproxies=nnproxy1,nnproxy2
    After: dfs.ha.namenodes.nnproxies=nn1,nn2

    1. Add the dfs.ha.namenodes.nnproxies=nn1,nn2 property.

    2. Ensure that the property has the Final option selected, so that the hdfs-site does not overwrite it.

    3. Save the configuration changes.

  3. Deploy the configuration and restart the Hive service.

Optimising performance

This section provides additional configuration properties that can be used to optimize the performance of NameNode Proxy.

Add the properties to the following locations depending on platform.

Ambari

NameNode ProxyConfigsCustom nn-proxy-server-site

Cloudera

NameNode ProxyConfigurationNameNode Proxy Server Advanced Configuration Snippet (Safety Valve) for proxy-site.xml (NameNode Proxy Server Default Group)

Property Name Description Default Value

npx.handler.count

The number of method handler threads to run. This is the number of threads running endlessly trying to fetch new RPC calls from a blocking queue on the RPC server and process them.

50

npx.reader.count

The number of reader threads to run. This is the number of threads running endlessly trying to read new RPC calls from connections and then putting them in to the blocking queue on the RPC server.

1

The NameNode Proxy service will need to be restarted afterwards to make the new configuration active.

Load Balancing

This configuration option has already been recommended in the Configuring NameNode Proxy HA sections (both Ambari and Cloudera), but is detailed here again for consideration.

Also note that it is only possible to implement this option if NameNode Proxy HA is configured.

The use of the dfs.client.failover.random.order.* property is only available in HDP 3 and CDH 6 onwards. If running a version lower than this, it is not possible to load balance the client requests using the method detailed below.
As such, proceed to the Fall-through section instead if running below HDP 3 and CDH 6.

When a client makes a request, the NameNode Proxy used can be randomised (if there is more than one). To configure this, set the following:

dfs.client.failover.random.order.<nnproxies-nameservice> = true

Once set, this configuration will help load balance client requests through any NameNode Proxy, rather than hitting the first NameNode Proxy host in the list.

It tells the clients that for the nameservice provided (<nnproxies-nameservice> in this example), the ConfiguredFailoverProxyProvider class should randomize the hosts contained within the nameservice specified.

Additional notes
  • If the dfs.client.failover.random.order.* property is not present, the default value is false.

  • Ensure the <nnproxies-nameservice> example is replaced by your desired nameservice.

  • There is also a dfs.client.failover.random.order configuration that makes all nameservices use a random order. This is potentially too invasive for other nameservices, as such, it is not recommended for NameNode Proxy.

  • If all NameNode Proxies are shut down, a request from the client will fail. To prevent this, use the Fall-through mechanism detailed below.

There is an alternative to the ConfiguredFailoverProxyProvider, which must not be used with the NameNode Proxy.
The RequestHedgingProxyProvider method (available in CDH 5.15.2/HDP 2.6.5 and above) may result in dual-execution of client requests as it implies all NameNode proxies are active. The standard use of this provider is for Active/Standby NameNodes, but as all NameNode proxies are technically active, this provider is not suitable.
Fall-through

Clients will try the NameNode Proxies in a deterministic order. This provides a failsafe if the NameNode Proxies were ever offline.

To configure the order, set the following:

  • dfs.client.failover.random.order.<nnproxies-nameservice> is set to false or is unset.

  • All configuration for <nnproxies-nameservice> contains all the NameNode Proxy hosts, followed by all the underlying NameNodes.

  • The downside of this configuration is that all client requests will be received by the same NameNode Proxy server (unless it is offline).

Known issue prior to Hadoop 2.9

HDFS-6648 is a bug where, prior to Hadoop 2.9, the configured list was not the order followed by ConfiguredFailoverProxyProvider. This is due to the ordering by the hash value of the host names. Therefore, prior to Hadoop 2.9, we can’t reliably order the NameNode proxies/hosts via configuration.

Router Based Federation

Due to the scalability limits of NameNodes, a common solution is to partition the filesystem into subclusters (e.g., namespace partitions).

Hadoop 2.9 and above has support for Router Based Federation (RBF). Router Based Federation (RBF) is a unified view of multiple subclusters that are configured via mountables. A Hadoop client interacts with the Router, rather than NameNodes, and the Router delegates the calls to the correct subcluster. It does so by looking at the mount table, where you can map a path to a certain namespace. NameNode Proxy has support of being a RBF subcluster as depicted in the below diagram:

NameNode Proxy RBF Subcluster
Figure 79. NameNode Proxy with Router Based Federation Subcluster

The above diagram shows that it is possible for NameNode Proxy to replicate the contents of the HDFS Cluster 1 subcluster, while ignoring HDFS Cluster 2 and 3.

To enable this, the following configuration changes are required:

This example uses the following variables:

NameNode Proxy nameservice = nnproxies
NameNode Proxy 1 unique identifier = nnproxy1
NameNode Proxy 2 unique identifier = nnproxy2
NameNode nameservice = nameservice01

Please adjust according to your environment.

  1. Add or edit the following properties in hdfs-site.xml via your Ambari Manager or Cloudera Manager.

    • Property = dfs.namenode.rpc-address.nnproxies.nnproxy1
      Description = RPC address of first NameNode Proxy.
      Example value = nnproxy1_hostname:8890

    • Property = dfs.namenode.rpc-address.nnproxies.nnproxy2
      Description = RPC address of second NameNode Proxy.
      Example value = nnproxy2_hostname:8890

    • Property = dfs.nameservice.id
      Description = The nameservice ID representing all NameNodes.
      Example value = nameservice01

    • Property = dfs.federation.router.monitor.namenode
      Description = Comma separated list of all NameNode Proxies unique identifiers.
      Example value = nnproxy1,nnproxy2

    • Property = dfs.client.failover.proxy.provider.nnproxies
      Description = Failover provider class.
      Example value = org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider

  2. After you have added or edited the above configuration, deploy the configuration and restart any affected services.

Note on Mount Tables

A Fusion instance can not service multiple name services and therefore can not support multiple mount tables that point to different name services.

After the configurations are set and the effected services have been restarted, mount tables can be created for NameNode Proxy’s nameservice.

hdfs dfsrouteradmin -ls /
Table 2. Example of Mount Table Entries
Source Destinations Owner Group Mode Quota/Usage

/nonrepl

othercluster→/directory

hdfs

hadoop

rwxr-xr-x

-/-, SsQuota: -/-

/repl

npxns→/repl1

hdfs

hadoop

rwxr-xr-x

-/-, SsQuota: -/-

In the above example, anything written to /repl will be routed to the NameNode Proxy nameservice, and, therefore, the NameNode Proxy.

If /repl1 is a replicated rule configured by Fusion, it will be replicated. Anything written to /nonrepl would go to the other cluster nameservice, which is a separate cluster not integrated with Fusion.

Client Distribution - Hortonworks

After WANdisco Fusion has been successfully installed, you can proceed to distribute and then configure the Fusion client.

  1. Fusion client packages need to be distributed before configuration is possible. Client packages for this server are available from the Cluster Interaction section of the Settings tab.

  2. Follow the on screen steps to download the Stack packages.

    wdf2.14.1 stacks
    Figure 80. Ambari - Client Distribution
  3. Open a terminal session to the location of your Stack repository, normally your Ambari Manager server. Ensure that you have suitable permissions for handling files.

  4. Download the appropriate stack for your deployment.

  5. Extract the Stack package and move it to your Ambari service directory. The service is a gz file (e.g. fusion-hdp-<your_version>.stack.tar.gz) that will expand to a directory called /FUSION.
    First check for and remove existing /FUSION directories in /var/lib/ambari-server/resources/stacks/HDP/<hdp-version>/services, before placing in the new /FUSION directory. If you don’t clear out a pre-existing directory you might end up installing an outdated version of the client.

  6. Restart the ambari-server

    ambari-server restart
  7. On the Ambari UI, after the server restarts, go to + Add Service.

  8. Scroll down the Choose Services window to varify that WANdisco Fusion is present on the list Do not add the service here, instead go back to the Fusion Installer UI and follow the on screen instructions.

  9. Click Deploy Stack.

    If you are using HDP 3.x and Kerberos, you also have to enter your KDC admin credentials. Once details are entered, validate and deploy the stack.

    wdf2.14 ambari client hdp3 only
    Figure 81. Ambari - Client Distribution
    Recovering from a stack deploy error
    If the deploy stack action fails, see this Knowledge base article for more information - Recovering from a stack deploy error.
  10. Return to the Steps required before Live replication section for guidance on publishing configuration and restarting services.

Using HttpFS
If you are using HttpFS, an extra step is required. Once the stack has been installed and HDFS has been restarted, the HttpFS service must also be restarted.
Client Distribution - Cloudera

After WANdisco Fusion has been successfully installed, you can proceed to distribute and then configure the Fusion client.

  1. Fusion client packages need to be distributed before configuration is possible. Client packages for this server are available from the Cluster Interaction section of the Settings tab.

  2. Follow the on screen steps to download the Parcel packages.

    WANdisco Fusion Client configuration
    Figure 82. Parcel Download
  3. Open a terminal session to the location of your parcels repository, normally your Cloudera Manager server. Ensure that you have suitable permissions for handling files.

  4. Download the appropriate parcel and sha for your deployment.

  5. Move the files into your Cloudera parcel directory, i.e.

    mv FUSION-*.parcel* /opt/cloudera/parcel-repo/
  6. Change the ownership of the parcel and .sha files so that they match the system account that runs Cloudera Manager:

    chown cloudera-scm:cloudera-scm FUSION-*.parcel*
  7. Open Cloudera Manager and navigate to the Parcels screen by clicking on the Parcel icon.

    WANdisco Fusion tree
    Figure 83. Open Cloudera Manager
  8. Click Check for New Parcels.

    WANdisco Fusion tree
    Figure 84. Check for new parcels
  9. The WANdisco Fusion client package is now ready to distribute. Distribute and then Activate.

  10. Return to the Steps required before Live replication section for guidance on publishing configuration and restarting services.

Using HttpFS
If you are using HttpFS, an extra step is required. Once the parcel has been installed and HDFS has been restarted, the HttpFS service must also be restarted.

6.1.8. MapR Configuration

These tasks will be required only for MapR deployments. Please ensure the Client Configuration is only performed after installing WANdisco Fusion.

MapR Client Configuration

On MapR clusters, you need to copy WANdisco Fusion configuration onto all other nodes in the cluster:

  1. Open a terminal to your WANdisco Fusion node.

  2. Navigate to /opt/mapr/hadoop/<hadoop-version>/etc/hadoop.

  3. Copy the core-site.xml and yarn-site.xml files to the same location on all other nodes in the cluster.

  4. Now restart HDFS, and any other service that indicates that a restart is required.

When using MapR and doing a TeraSort run, if one runs without the simple partitioner configuration, then the YARN containers will fail with a Fusion Client ClassNotFoundException. The remedy is to set yarn.application.classpath on each node’s yarn-site.xml.
MapR Impersonation
Enable impersonation when cluster security is disabled

Follow these steps on the client to configure impersonation without enabling cluster security.

  1. Enable impersonation for all relevant components in your ecosystem. See the MapR documentation - Component Requirements for Impersonation.

  2. Enable impersonation for the MapR core components:
    The following steps will ensure that MapR will have the necessary permissions on your Hadoop cluster:

    • Open the core-site.xml file in a suitable editor.

    • Add the following *hadoop.proxyuser* properties:

      <property>
          <name>hadoop.proxyuser.mapr.hosts</name>
          <value>*</value>
      </property>
      <property>
          <name>hadoop.proxyuser.mapr.groups</name>
          <value>*</value>
      </property>

      Note: The wildcard asterisk * lets the "mapr" user connect from any host and impersonate any user in any group.

    • Check that your settings are correct, save and close the core-site.xml file.

  3. On each client system on which you need to run impersonation:

    • Set a MAPR_IMPERSONATION_ENABLED environment variable with the value, true. This value must be set in the environment of any process you start that does impersonation. E.g.

      export MAPR_IMPERSONATION_ENABLED=true
    • Create a file in /opt/mapr/conf/proxy/ that has the name of the mapr superuser. The default file name would be mapr. To verify the superuser name, check the mapr.daemon.user= line in the /opt/mapr/conf/daemon.conf file on a MapR cluster server.

6.1.9. Client Bypass

Client bypass to allow writes to proceed

If WANdisco Fusion is down and clients use the HDFS URI, then further writes will be blocked. The client bypass feature gives the administrator an option to bypass WANdisco Fusion and write to the underlying file system, which will introduce inconsistencies between zones. This is suitable for when short-term inconsistency is seen as a lesser evil compared to blocked progress.

The inconsistencies can then be fixed later using the Consistency and Make Consistent process(es). A client that is allowed to bypass to the underlying filesystem will continue to bypass for the duration of the retry interval. Long-running clients will automatically reload configurations at a hardcoded 60 second interval. Thus it is possible to disable and enable the bypass on-the-fly.

Enable/disable client bypass via the UI
  1. Log in to the Fusion UI and go to the Settings tab. Click Client Bypass Settings.

    Enable Client Bypass
    Figure 85. Client Bypass - step1
  2. Tick the Enable fusion bypass checkbox. This will enable two entry fields for configuration:

    Enable Client Bypass
    Figure 86. Client Bypass - step2
    Bypass response time

    The time (in seconds) that will pass before the client will bypass WANdisco Fusion. Default: 14.

    Bypass retry interval

    The time (in seconds) before the client attempts to use WANdisco Fusion, again. Default: 60.

  3. Click Update to save your changes.

  4. A HDFS restart is now needed for the changes to take effect.

Enable/disable client bypass via manual configuration change

In core-site.xml add the following properties:

  • fusion.client.can.bypass = true

    • default is false

  • fusion.client.bypass.retry.interval.secs = 120

    • default is 60 (in seconds)

The properties are also listed in the Fusion Client configuration table.

6.2. Induction

Induction is the process used to incorporate new nodes/zones into the WANdisco Fusion LiveData platform.

Use this procedure to induct two zones together in order to create a replication ecosystem.

Known issue - HTTP authentication must be disabled when inducting to a new zone or adding to an existing zone

Before adding/inducting, disable HTTP authentication via Fusion Kerberos on the Settings page.
Once the nodes are inducted, HTTP authentication can be enabled again.

  1. Log in to one of the active nodes, clicking on the Nodes tab. Click the + Induct button.

    WANdisco Fusion Deployment
    Figure 87. Induct node
  2. Enter the FQDN of the new node that you wish to induct and the Fusion Server Port. Click Start Induction.

    WANdisco Fusion Deployment
    Figure 88. Remote node details
    Fully Qualified Domain Name

    The full domain name for the new node that you will induct into your replication system.

    Fusion Server Port

    The TCP port used by the WANdisco Fusion application for configuration and reporting, both internally and via REST API. The port needs to be open between all WANdisco Fusion nodes and any systems or scripts that interface with WANdisco Fusion through the REST API. Default is 8082.

  3. When the induction process completes you have the option to induct another node. The Nodes tab will refresh with the new node added to the list.

    WANdisco Fusion Deployment
    Figure 89. Induction complete
Re-make Replication Rules
To add your newly inducted nodes to existing Replication Rules you will need to manually remove existing rules and create new ones including the new node. This process will be improved in future releases.
For information on how to create rules see Create a Rule.

6.2.1. Induction Failure

The induction process performs some validation before running. If this validation failures you will quickly see a warning messages appear.

WANdisco Fusion Deployment
Figure 90. Induction failure
Automatic Induction Failure

If the induction process can’t connect to the new node using the details provided, a failure will happen instantly. This could happen because of an error in the new node’s installation, however it could also be caused by the node being kerberized.

We also could not reach any of our standard ports

If connections can’t be made on specific Fusion ports, they will be listed here. If none of the standard ports are reachable then you will be warned that this is the case.

Fully Qualified Domain Name

The full hostname for the server.

Node ID

A unique identifier that will be used by WANdisco Fusion UI to identify the server.

Location ID

This is the unique string (e.g. "db92a062-10ba-11e6-9df2-4ad1c6ce8e05") that appears on the Node screen (see below).

Fusion Port

The TCP port used by the replication system. It needs to be open between all WANdisco Fusion nodes. Nodes that are situated in zones that are external to the data center’s network will require unidirectional access through the firewall.

6.3. In-life configuration

This section covers in-life configuration options available, some of which will be performed in the Fusion UI.

Many of the options within the Settings page will have already been covered in the Initial configuration section, as they are required for the initial setup of Fusion.

6.3.1. About This Node

This panel will display information regarding the version of Fusion installed, as well as uptime of the server and the environment type (e.g. AMBARI, CLOUDERA, UNMANAGED S3).

About - node name
Figure 91. About this node
Node name

To change the Node name value, enter a different string in the editable box and select Update. This change will then propogate to other nodes in the ecosystem after a short period.

6.3.2. Fusion Server settings

The Fusion Server settings allow you to adjust Java heap assigned to the server process, URI Selection, and some other additional options.

Fusion Server Settings
Figure 92. Fusion Server Settings
Maximum Java heap size (GB)

Enter the maximum Java heap value for the Fusion Server process. The minimum for production is 16GB but 64GB is recommended.

DConE panic if database is dirty (checkbox)

This option lets you enable the strict recovery option for WANdisco’s replication engine, to ensure that any corruption to its prevayler database doesn’t lead to further problems. When the checkbox is ticked, the Fusion Server will log a panic message whenever it is not properly shutdown, either due to a system or application problem.

Application panic if database is dirty (checkbox)

This option lets you enable the strict recovery option for Fusion Server’s database, to ensure that any corruption to its internal database doesn’t lead to further problems. When the checkbox is ticked, the Fusion Server will log a panic message whenever it is not properly shutdown, either due to a system or application problem.

URI Selection

For detail and explanation on each option, please refer to the On-premises - Fusion options for URI selection or Cloud - Fusion options for URI selection.

If considering a change to the URI selection, contact WANdisco support prior to making any such change.

Please note that any change to this option will update the core-site potentially requiring restarts of services. Depending on the option selected, it may also require packages to be distributed and others removed. This should be carefully planned to avoid outages or unexpected behaviour.

Set Push Threshold Manually

The feature exposes the configuration property fs.fusion.push.threshold, stored in the core-site.xml file (see Reference - core-site.xml section for more information).

It provides administrators with a means of making a small performance improvement, useful in a small number of cases.

Fusion Settings - push threshold
Figure 93. Set Push Threshold manually

When enabling the checkbox in the UI, the entry displays as "Required". You can enter your own value (in bytes) and click the Update button.

Amazon cloud deployments
For cloud deployments, ensure the property is disabled (unticked) or set to zero "0". This will disable HFLUSH which is suitable because appends are not supported for S3 storage.
Chunk Size in Bytes

The size of the 'chunks' used in a file transfer between Zones (in Fusion 2.14, the default value is 32768).

The property may be adjusted depending on the environment and expected file sizes that will be in transit. If expecting many large files to be transferred and a decent bandwidth is in place between Zones, it may be advantageous to increase this value for performance gains. For Production environments, please seek guidance from WANdisco support before adjusting this value.

To adjust, enter your own value (in bytes) in the editable field and click the Update button.

6.3.3. Graphs

The severity percentage thresholds for the graphs displayed on the Fusion UI Dashboard can be adjusted.

Graph Settings
Figure 94. Graph Settings

Edit the values as required and select to Update. The changes will be reflected on the Dashboard page shortly afterwards.

Dashboard graphs
Figure 95. Dashboard - Graphs

6.3.4. IHC Server Settings

This section will display the configurable options available for the IHC Server.

IHC Server Settings
Figure 96. IHC Server Settings
Maximum Java heap size (GB)

Enter the maximum Java heap value for the IHC Server process. The minimum recommendation for production is 16GB.

6.3.5. License Settings

Unless there’s a problem that stops you from reaching the WANdisco Fusion UI, the correct way to upgrade a node license is through the License Settings panel, under the Settings tab.

License Settings
Figure 97. License Settings
  1. Click Select a file. Navigate to and select your replacement license file.

  2. Click Upload and review the details of your replacement license file.

    License Settings - validation
    Figure 98. License Settings - validation
  3. Once the License Validation is complete, select to Update.

  4. A warning message will be displayed:

    Any Fusion Nodes which are currently down will not receive this license update. To update the license on these nodes a manual update will be required, or this license update can be run again when they are back online. Are you sure you want to continue with the License Update now?

    Click Confirm once the message is read and understood.

After the license upload is complete, you will see an additional message:

License updated but will not be visible on the UI until the page is refreshed. Note that the WANdisco Fusion Server only checks for new licenses once an hour. You can force a check by restarting the WANdisco Fusion server.

License expiry/renewals

If a license is due to expire soon, the Fusion UI provides a warning message on the Dashboard.

license
Figure 99. License expiry warning

When uploading a new license as described in License Settings, this warning will remain until the License check is performed automatically every hour. It is possible to remove this warning immediately by restarting the Fusion UI (service fusion-ui-server restart).

There is also an email address provided on the Support tab of the Fusion UI for requesting new licenses.

license
Figure 100. Support - Purchase a License

Alternatively, contact WANdisco support if there are any questions or concerns regarding licenses.

Manual license update

The following manual procedure should only be used if the Fusion UI is not accessible, such as when a node cannot be started - maybe caused by ownership or permissions errors on an existing license file.

  1. Open a terminal session to the Fusion node, and navigate to the Fusion server properties directory:

    cd /etc/wandisco/fusion/server
  2. We recommend that you rename the license.key to something versioned, e.g. license.20190715.

  3. Upload the new license key to the /etc/wandisco/fusion/server directory. You need to account for the following factors:

    1. Ensure the filename is license.key.

    2. Ownership should be the same as the original file.

    3. Permissions should be the same as the original file.

  4. Restart the Fusion server by running the Fusion init.d script with the following argument:

    service fusion-server restart

    This will force WANdisco Fusion to pick up the new license file and apply any changes to permitted usage.

    If you don’t restart
    If you follow the above instructions but don’t do the restart WANdisco Fusion will continue to run with the old license until it performs a daily license validation (which runs at midnight). Providing that your new license key file is valid and has been put in the right place then WANdisco Fusion will then update its license properties without the need to restart.
  5. Ensure that this task is performed on all Fusion nodes in the ecosystem.

  6. If you run into problems, check the Fusion server logs (/var/log/fusion/server/) for more information and contact WANdisco support.

    Example error

    PANIC: License is invalid com.wandisco.fsfs.licensing.LicenseException: Failed to load /etc/wandisco/fusion/server/license.key

6.3.6. Node: Log Settings

Unless configured differently, WANdisco Fusion logs default to the following directories:

  • Fusion Server logs - /var/log/fusion/server

  • IHC Server logs - /var/log/fusion/ihc

  • Fusion UI server logs - /var/log/fusion/ui

Log Settings
Figure 101. Log Settings

The logging directories for each component can be changed in the editable fields. Before changing the directory, you must ensure that the new directory exists and that the Fusion system user has full permissions on it. Select to Update after any changes have been made.

These changes will only take effect after restarting the Fusion UI Server. This should be performed in a terminal session on the Fusion node.

service fusion-ui-server restart
Changes via the command line

To change the logging directory, you can also edit the /etc/init.d/fusion-ui-server script on the Fusion node. The property to edit depends on the log you wish to change.
Example
export LOG_DIR="/var/log/fusion/ui".

You need to restart the Fusion UI Server for changes to take effect.

service fusion-ui-server restart

6.3.7. Plugins

This panel will display any plugins installed on the Fusion server, and the current status of them.

Settings - Plugins
Figure 102. Plugins

Whilst there is generally nothing to configure on this page, when installing a new plugin, there will often be an option to Install now. This is used to facilitate the UI driven aspects of the plugin installation journey, not all plugins will require this though.

If you have any concerns regarding the status of a plugin shown, contact WANdisco support.

6.3.8. UI Settings

You can change how you interact with the Fusion UI through the browser.

UI Settings
Figure 103. UI Settings
HTTP UI Port

The port assigned to the Fusion UI.

HTTP UI Host

The host that the Fusion UI is set to, by default, this will be 0.0.0.0 (i.e. it will bind to the server’s default route).

HTTP UI External Host

This is the external hostname defined for the server. It is used in the Fusion UI for the Jump to node mechanism seen in the Nodes page. If adjusted, it will redirect the browser to the hostname specified in this field as the change will be broadcasted to all other nodes in the ecosystem.

Nodes - Jump to node
Figure 104. Nodes - Jump to node

Click Update after making any changes. A restart of the Fusion UI will also be required, this should be performed in a terminal session on the server:

service fusion-ui-server restart

You may need to update the URL in your browser to account for any changes made.

If wanting to Use HTTPS on the UI, or guidance on the Trust Store Settings, see the Enable SSL (HTTPS) for the Fusion UI section.

6.3.9. Consistency Check - Scheduled

This section allows you to enable or disable scheduled consistency checks, alter the default check interval and change the consistency check report limit.

To learn about running Consistency Checks on replication rules, proceed to the Consistency Check section in Replication.
Consistency Check
Figure 105. Schedule consistency check
Enable Scheduled Consistency Checks

This determines if scheduled consistency checks are enabled. Altering this can be applied to existing rules and/or newly created rules.

Update

Click Update after enabling or disabling this feature.

Apply to existing rules

This applies the enablement/disablement to all replication rules, except those with their own schedule.

Default Check Interval

This can be set to repeat hourly, daily or weekly. Hourly lets you set the default number of hours (1-12) that are allowed to pass before the replication rules are checked for consistency. Daily lets you set the time of day for when the consistency checks are performed. Weekly lets you set the day of the week and time of day that the consistency checks will occur.

Update

Click Update to store the entered value and use it for all new replication rules that do not have their own set interval.

Apply to existing rules

This applies the new default value to all replication rules, except those with their own schedule.

Consistency Check Report Limit

The maximum number of inconsistencies to be displayed in reports on the Fusion UI.

It’s possible to set a different schedule for each specific replication rule using the Advanced Options available when setting up or editing a Replication Rule.
See Create a Rule or View/Edit Replication Rule for guidance.

6.3.10. Hadoop: Log Settings

This section details options available for Hadoop logging (this will not be available on unmanaged platforms).

Hadoop - log settings
Figure 106. Hadoop - Log Settings
Fusion Client Logging
Show log statements for Fusion Client in Hadoop console

If logging is enabled, then all log statements of the Fusion Client will be shown when Hadoop applications and users interact with HDFS.

This will only be pertinent if the Fusion Client is being called, such as if the HDFS URI with HDFS URI scheme is being used on the cluster.

6.3.11. Fusion Administrator

This section can be used to activate the Fusion Administrator for the Fusion UI, as well as change the username and password assigned to it.

UI Settings - Fusion Administrator
Figure 107. UI Settings - Fusion Administrator

The Fusion Administrator and any user whose role is mapped to the Fusion Administrator group can enable LDAP and configure the necessary Fusion roles. This user is set during the Security step of the installation of Fusion.

Activate Fusion Administrator

This is activated by default, by deactivating it, you must ensure that the Cluster Manager details have been set or else it may not be possible to log into the Fusion UI again without manual intervention.

Note that enablement of LDAP is only possible using the Fusion Administrator.

User Name

Set the User name for the Fusion Administrator.

User Password

Set the password for the Fusion Administrator.

Confirm User Password

Confirm the password for the Fusion Administrator.

Select to Update once any changes have been made. You will then need to enter the new details when relogging into the Fusion UI.

6.3.12. LDAP Settings

Use the following procedure for setting up and applying an LDAP/AD service for handling Fusion user authentication.

OpenLDAP and Active Directory implementations are both supported.
  1. Login using the Fusion Administrator credentials that you provided during installation (see Fusion Administrator section for details). Navigate to the Settings tab and click on LDAP under the Security Section.

    Use LDAP

    Tick the checkbox if wanting to enable LDAP authentication. The following properties will then be displayed.

    LDAP Settings
    Figure 108. LDAP Settings - Use LDAP checked
    Host

    URL or hostname for your LDAP/AD authority

    Port

    Port used by your LDAP/AD authority. Default: 389.

    Use SSL

    Tick this checkbox if you wish to secure LDAP/AD traffic using SSL. The Port may need to be adjusted to one that is SSL enabled on the LDAP/AD host.

    LDAP Settings - Use SSL
    Figure 109. LDAP Settings - Use SSL
    TrustStore Path

    Absolute file path to your SSL TrustStore file.

    TrustStore Password

    Password required for your TrustStore.

  2. Enter the LDAP Authority details.

    BindDN

    The Distinguishing Name is used by Fusion to authenticate to LDAP/AD in order to read groups, etc. In the case that your authority does not allow unauthenticated queries.

    Bind Password

    Password required to query your LDAP/AD authority.

  3. Enter the Search details for the chosen group of users.

    User Search Base

    Point in the LDAP/AD tree where the search for valid users will begin.

    User Search Filter

    LDAP/AD search string used to filter the directory for applicable accounts.

    Login Name Attribute

    Name of the UID attribute for users in the directory.

  4. Click Check connection. The following message will be displayed if a connection was successful:

    Settings are valid. A connection to the LDAP server could be created.

  5. Click Preview Users. A list of users will be displayed, ensure that this is the expected user group.

  6. If your LDAP/AD settings are confirmed as valid, and the user group is correct, then you must click Update to save the configuration. If you don’t click Update then they won’t be saved.

  7. After clicking Update a message will appear to confirm that the LDAP settings will now be applied.

    LDAP settings successfully updated. Please reload the browser.

    Click Close and refresh your browser.

6.3.13. Email Notifications

This section describes how to set up notification emails that will be triggered if one of the tracked system resources reaches a defined threshold.

Email notification is disabled by default.
You must complete the following steps before any messages will be sent.

Email Notification Settings are located in the Zone section of the settings menu.

Complete the following steps to enable email notification:

  1. Enter your SMTP properties in the Server configuration tab.

  2. Enter recipient addresses in the Recipients tab.

  3. Tick the Enable check-box for each trigger-event for which you want an email notification sent out.

  4. [Optionally] You can customize the messaging that will be included in the notification email message by adding your own text in the Templates tab.

Notification email triggers

The following triggers support email notification. See the Templates section for more information.

Consistency Check Failing

Email sent if a consistency check fails.

CPU Load Threshold Hit

Dashboard graph for CPU Load has reached. See Dashboard Graphs Settings for adjusting thresholds.

Fusion Database Partition Disk Threshold Hit

Dashboard graph for Database partition disk usage has been reached. See Dashboard Graphs Settings for adjusting thresholds.

License Data Threshold Hit

Licensed data usage threshold has been reached. See Dashboard Graphs Settings for adjusting thresholds.

License Expiring

The Node’s WANdisco Fusion license is going to expire.

Node Down

One of the Nodes in your ecosystem is down.

Quorum Lost

One or more of the Zones are unable to continue replication due to the loss of one or more nodes.

Server config

The server config tab contains the settings for the SMTP email server that you will use for relaying your notification emails. You need to complete and check the provided details are correct first, before your notification emails can be enabled.

SMTPSettings
Figure 110. SMTP settings

Email Notification Settings are located in the Zone section of the Settings tab

SMTP Host

The hostname or IP address for your email relay server.

SMTP Port

The port used by your email relay service. SMTP default port is 25.

Connection Encryption

Drop-down for choosing the type of encryption that the mail server uses, None, SSL or TLS are supported. If SSL or TLS are selected you should make sure that you adjust the SMTP port value, if required.

Authentication

Checkbox for indicating that a username and password are required for connecting to the mail server. If you tick the checkbox additional entry fields will appear.

SMTP Username

A username for connecting to the email server.

SMTP Password

A password for connecting to the email server.

From

Optional field for adding the sender email address that will be seen by to the recipient.

To

Optional field for entering an email address that can be used for testing that the email setup will work.

Update Settings

Button, click to store your email notification entries.

Reset Changes

Reloads the saved settings, undoing any changes that you have made in the template that have not been saved.

Send Test Email

Trigger the sending of a test email to confirm that it works as expected.

Recipients

The recipients tab is used to store one or more email addresses that can be used when sending out notification emails. You can enter any number of addresses, although you will still need to associate an entered address with a specific notification before it will be used. See Adding recipients.

Email Notification Settings
Figure 111. Adding recipients
Adding recipients
  1. Enter a valid email address for a recipient who should receive a notification email from WANdisco Fusion.

  2. Click the Add button.
    You can repeat the procedure as many times as you like, you can send each different notification to a different recipient (by associating that recipient’s address with the particular trigger), or you can send a single notification email to multiple recipients (by associating multiple addresses with the notification email.

  3. Click Update Recipients to save the new email addresses.

Enable Notification Emails

Once you have working server settings valid recipient email addresses you can start to enable notification emails from the Alerts tab.

  1. Go to the Alerts tab and select a notification trigger for which you would like to send emails. For example Consistency Check Failing. Tick the Enabled checkbox.

    If a trigger is not enabled, no email notification will ever be sent. Likewise, an enabled trigger will not send out notification emails unless recipients are added.
    Email Notification
    Figure 112. Email Notification Enabled
  2. Enter a From email address and select recipients from the Add More Recipients window. Once you have finished selecting recipients, click Add.

    Email Notification
    Figure 113. Email Notification - Add
  3. Scroll to the bottom of the screen and click Update Alerts to save changes.

  4. You can choose to change/add additional recipients, review or customize the messaging by clicking on the Edit Template link.

Templates

The Templates tab gives you access to the email default text, allowing you to review and customize with additional messaging.

The types of template available are:

  • Consistency Check Failing

  • CPU Load Threshold Hit

  • Fusion Database Partition Disk Threshold Hit

  • License Data Threshold Hit

  • License Expiring

  • Node Down

  • Quorum Lost

EmailSettings
Figure 114. Email Settings
Consistency Check Failing

This is the trigger system event for which the notification email will be sent.

Subject

The email’s subject line. A default value is set for each of the triggers, however, you can reword these by changing the text in the template.

Custom Message

This entry box lets you add your own messaging to the notification. This could be anything that might be useful to an on-duty administrator such as links to related documentation or contact details for the next level of support, etc.

Message Body

The message body contains the fixed payload of the notification email; you can’t edit this element and it may contain specific error messaging taken from logs.

The subject and custom messages can also be altered in the file /opt/wandisco/fusion-ui-server/properties/email.properties. This allows, for example, the email subject to be in the language of the recipients.

Example Notification Email

This is what an email notification looks like:

    From: cluster-admin@organization.com>
    Date: Mon, Nov 6, 2017 at 3:49 PM
    Subject: WANdisco Fusion UI - Consistency Check Failing
    To: admin@company.com

    Here is a custom message.
     - Custom messaging entered in the Template

    Consistency Check Failing triggered a watch event, any relevant error message will appear below.
     - Default Message

    The following directory failed consistency check:

      /repl1
    - Specific error message

    ==================== NODE DETAILS =====================
    Host Name     : xwstest-01.your.company.com
    IP address    : 10.0.0.146
    IP port       : 6444
    -------------------------------------------------------
    Node Id       : f5255a0b-bcfc-40c0-b2a7-64546f571f2a
    Node Name     : wdfs1
    Node status   : LOCAL
    Node's zone   : zone1
    Node location : location1
    Node latitude : 11.0
    Node longitude: 119.0
    -------------------------------------------------------
    Memory usage  : 0.0%
    Disk usage    : 0.0%
    Last update   : 2017.Nov.06 at 15:49:28 GMT
    Time Now      : 2017.Nov.06 at 15:49:48 GMT
    =======================================================
     - Standard footer

6.3.14. Networking settings

This section lets you define whether or not the WANdisco Fusion server proactively creates outbound connections to remote IHC servers during data transfer. It will otherwise tell the server to wait for and re-use inbound connections.

Setting to Inbound connection on a Source zone is useful if, for example, the source datacenter has networking restrictions preventing multiple ports being made available to the destination zone(s).

When set to Inbound, the WANdisco Fusion Server must have a hostname visible to other Zones that will be inducted to this ecosystem.
Outbound connection

The default network setting (Networking direction between Fusion Server and IHC Server) is set as Outbound connection, on the Networking Panel under the Settings tab.

WANdisco Fusion tree
Figure 115. unidirectional networking - Outbound
Inbound connection

When Inbound connection is selected, then you must ensure that Fusion server must be publicly visible to other zones. To ensure this, you must enter a Fusion Server/Local IHC Public IP Address.

Networking settings - inbound
Figure 116. Unidirectional networking - Inbound
Fusion Server/Local IHC Public IP Address

Changing the hostname will require a restart of both the Fusion and IHC server, in order for the change to take effect.

Hostname cannot be changed after induction
Note that once a WANdisco Fusion node has been inducted, it is no longer possible to change its hostname.
Remote IHC Port

The port that remote IHC servers will use to contact this Fusion server in the case of Inbound connections.

6.3.15. Username Translation Setup

It is possible to setup username translation between zones during replication of files and directories. See the Username Translations section for further information about this feature.

Adding username translation will cause usernames to be ignored in consistency checks.
  1. Go to the Settings tab and select Username translation from the Zone section.
    As any changes made in the UI are replicated to all nodes and zones, you can be, for example, on a node in zone 3 to create a mapping between zone 1 and 2.

    wdf2.14 username translation 01
    Figure 117. Select Username Translation on the Settings tab
  2. Select the zone from the drop down list, and enter the username for the source. Usernames can be defined explicitly or as a regex.

  3. Select the zone from the drop down list, and enter the username for the destination. Click Add.

    wdf2.14 username translation 02
    Figure 118. Enter zone and username details

    If the translation clashes with an existing one you will be warned at this point and the mapping will not be created.
    Note that mappings only exist as a one-way relationship and so you will need to add both directions if required.

  4. When all translations have been successfully added, click Update.

6.3.16. Replicating using the Filesystem’s scheme

Any WANdisco Fusion supported FileSystem can replicate using its own scheme, the fusion scheme does not have to be used. This allows you to have some schemes replicated using WANdisco Fusion and others independent of WANdisco Fusion. Creating your own schemes lets you, for example, change over time which sets of files are replicated. This may be useful if when setting up WANdisco Fusion you do not wish to replicate all files initially.

You can configure this with the following steps:

  1. Set fs.<scheme>.impl to com.wandisco.fs.client.FusionHcfs in the core-site.xml file. The <scheme> can have any name but it is advisable for it to be logical.

  2. If the <scheme> is not one of the ones provided by Fusion, update fs.fusion.underlyingFsClass. The value needs to be the full filesystem class, for example org.apache.hadoop.fs.RawLocalFileSystem.
    The schemes provided are listed in the Fusion Client Configuration section.

  3. Now deploy the client configs.

The result of this is that the fs.<scheme>.impl now points to Fusion, and Fusion interacts with the original FileSystem. Any client side application using the <scheme> will first go through Fusion.

Certain products make assumptions about the scheme used, for example Sentry ignores everything which is non-hdfs.

More information can be found in the Fusion Client Configuration section.

6.3.17. Configure for High Availability Hadoop

If you have configured High Availability (HA) on your Hadoop clusters since installing WANdisco Fusion you need to complete the following steps.

  1. Edit the Fusion configuration element fs.fusion.underlyingFs to match the new nameservice ID in the cluster-wide core-site.xml in your Hadoop manager.
    e.g. change:

    <property>
            <name>fs.fusion.underlyingFs</name>
            <value>hdfs://vmhost08-vm0.cfe.domain.com:8020</value>
    </property>

    To:

    <property>
            <name>fs.fusion.underlyingFs</name>
            <value>hdfs://myCluster</value>
    </property>
  2. Click Save Changes.

  3. Restart all Fusion and IHC servers once the client configurations have been deployed.

6.4. Replication

6.4.1. Setting up Replication

The following steps can be used to set up replication of data. The detail of each step will depend on your cluster setup and your specific replication requirements, although the basic steps remain the same.

  1. Create and configure a Replication Rule. See Create a Replication Rule.

  2. Perform a consistency check on your replication rule. See Consistency Check.

  3. Perform a Make Consistent operation on your replication rule. See Make Consistent.

  4. Run tests to validate that your replication rule remains consistent between Zones whilst data is being written to each data center.

6.4.2. Managing Replication

Replication tab

The Replication tab contains the Replicated Rules table.

Replication Tab
Figure 119. Replication Tab

This table lists all available rules and gives basic information:

Type

HCFS is the only rule type available with the core Fusion product but additional plugins are available to extend functionality. For example Hive rules are available with the Fusion Plugin for Live Hive.

Resource

The name give to the Replication rule.

Zone

The list of Zones that the rule has been assigned to. Hovering over the ? next to each Zone name will display the underlying filesystem path for the replication rule in that Zone. These can differ between Zones if the path mapping feature has been utilised.

Status

The summary of the latest Consistency Check, if one has occurred.

Activity

Basic information on the files being transferred.

From the top bar of the table you can create and remove rules.

Use the checkbox next to the rule (left column) to select it. You can then run make consistent and consistency check operations on the rule.

The filter option is also available to display specific rules by their type, resource, status, consistency or zone.

Clicking on the rule in the table takes you to the Replication Rule profile screen, giving more details about the rule. See the section on View/Edit Replication Rule for more information.

Filtering
Replication rules - filter
Figure 120. Replication rules - filter

In deployments that use large numbers of rules, you can use the filter tool to focus on specific sets of rules, filtering by type, resource, status, consistency or zone.

Filtering by Resource uses a text box, the other options provide a drop down list of options to filter by.

Create Replication Rules
  1. Log in to Fusion’s UI on one of the Fusion nodes and click on the Replication tab.

  2. Click on the + Create button. The Replication Rule form screen will appear.

    Create rules
    Figure 121. Create new rule(s)
  3. Zones
    Select two or more zones from the Zones list. The local Zone for the Fusion UI that is being used will always be selected. If only two Zones exist in the ecosystem, the second Zone will also be selected by default.

  4. Priority Zone
    The zone which is deemed most important, in terms of the most valuable/reliable source of data. For example, if your set up has a production zone and a disaster recovery zone, the production zone would be more important and, therefore, the priority zone.

  5. Rule Name
    Define a Rule Name for the replication rule, this must be a unique name and must not match one that is already created (a warning will be displayed if the name matches another rule).

  6. Path for <Zone>
    Define the path for replication. This can be done by either entering the path manually in the Path for <zone> field, or by using the HDFS File Tree on the right-hand side of the Rules panel to select your path.

    The grey folder icon next to the name of the path in the HDFS File Tree can be clicked to display any sub-directories within it. Clicking on the path name itself (e.g. tmp) will add this path to all Path for <Zone> fields.

    Path Mapping

    From Fusion version 2.14 onwards, it is possible to configure a replicated path that has different destination paths for a remote Zone. Simply adjust the Path for <Zone> fields for the Zones that require a different destination path.

    path mapping
    Figure 122. Path mapping creation

    Note that the HDFS File Tree will only display the local Zone’s filesystem, and the remote Zone(s) path(s) must be entered manually if they are to differ (please ensure they are correct before creating the rule).

    Path mapping cannot be altered after the replication rule has been created. If a change is required, the rule will need to be removed and created again.

    Once complete, click Add.

  7. You can create multiple replication rules at once by following the same procedure again. If you need to remove a rule, just click the x next to it in the Rules to be created list.

    Rules to be created
    Figure 123. Rules to be created
  8. You can now complete the creation of the New Rule(s) by clicking on the Create rules (N) button. However, there are some additional options available on the Advanced Options panel that you may need to configure. These are detailed in the following section.

Advanced Options - Replication Rules

Click on Advanced Options to display the additional options.

rule
Figure 124. Create rule - advanced options
For non-cloud based deployments
Shared Encrypted KMS

In deployments where multiple zones share a command Key Management Service (KMS), then enable this parameter to specify a virtual prefix path.

Preserve Origin Block Size

The option to preserve the block size from the originating file system is required when Hadoop has been set up to use a columnar storage solution such as Apache Parquet (this is common for Hive related data). If you are using a columnar storage format in any of your applications then you should enable this option to ensure that each file sits within the same HDFS block.

Preserve Replication Factor

By default, data that is shared between clusters will follow the local cluster’s replication rules rather than preserve the replication rules of the originating cluster. When this option is enabled, the replication factor of the originating cluster is preserved.

For all deployments
Schedule Consistency Checks

Select this option to set a consistency check interval that is specific to the rule and overrides the default value set in the Consistency Check section of the Settings tab. The consistency check can be set hourly, weekly or daily.

Exclude from replication

This lets you set an exclude pattern to indicate files and directories in your Replication Rules that are not to be replicated. See Exclude from replication for more information.

Ignored Zones

By using Ignored Zones, it is possible to block all write operations from one Zone to another. Select the applicable Zones to create a unidirectional restriction from one Zone to another. See Ignore Zones for a replicated rule for more information.

Exclude from replication

In the Advanced Options for a replication rule, exclusion rules can be defined for files and/or directories. These exclusion rules will prevent specified files or directories from being replicated between Zones. They will also be ignored during consistency checks so that only relevant comparisons are made between filesystems/object stores.

This feature can be used to prevent the replication of temporary system files that may not be required for replication (such as the Hive staging directories). The entry field will accept specific directories and files, or a glob pattern (text with wildcard (*) characters) that will match a set of directories and/or files.

Path interpretation
If the path contains a leading slash (/), Fusion will assume it is an absolute path. If it contains no leading slash then Fusion will assume it is a relative path and the root directory will be added to the beginning of the exclusion.
Default Exclusions

The following glob patterns are automatically excluded from replication:

/**/.fusion
/**/.fusion/**
  • These directories store Fusion’s housekeeping files, it should always be excluded in the global zone properties (even after an upgrade). It contains data relevant to the local nodes in the Zone.

/**/.Trash
/**/.Trash/**
  • Trash directories are excluded by default but can be removed if required.

/**/.hive-staging**
  • Hive staging directories are created temporarily as part of normal Hive operations, therefore, they are excluded from replication by default.

Example of a custom exclusion
Requirement

Exclude all files in the replicated directory with the norep_ prefix.

Directory structure
/repl1/rep_a
/repl1/norep_b
/repl1/subdirectory/rep_c
/repl1/subdirectory1/norep_d
/repl1/subdirectory2/rep_e
/repl1/subdirectory2/norep_e
Required rule
**/norep_*
  • Exclusion patterns do not need to reference an absolute path. Patterns are automatically treated as relative to the replication rule.

    Example
    Do not use /repl1/subdirectory2/no_rep_3, instead use /subdirectory/no_rep_3.

  • Take care when adding exclusion rules as there is no validation contained within the Fusion UI.

Ignore Zones for a replicated rule

From Fusion version 2.14, it is now possible to define an ignored zone for a replicated path in the Fusion UI.

By using Ignored Zones, it is possible to block all write operations (such as deletes) from one Zone to another. This includes both Live replication and Make consistent operations.

A common use-case for this feature would be to block all incoming write operations from a Disaster Recovery (DR) Zone to a Production (Prod) Zone. If a user performs a delete of a file in the DR zone or initiates a Make consistent operation with the DR zone as the source of truth, the Ignored zone restriction would prevent any changes on the Production zone.

Prior to 2.14, it was not possible to complete Consistency Checks on a replication rule when an ignored zone was included. This is no longer the case and consistency checks will now run as normal.
Setting an Ignored Zone in a replication rule

In Advanced Options, when creating a rule or editing an existing one, there is an option to set a unidirectional (one-way) restriction from one Zone to another.

Ignore Zones
Figure 125. Ignored Zones

The Zone that is set to be ignoring is selected on the left, and the Zone to be ignored is selected on the right. Once the Zone relationship is defined, Add the restriction.

Ignore Zones filled
Figure 126. Ignored Zones example

Select to Update if it is an existing rule, or Create rules (N) if it is new (and if everything else is complete).

View/Edit Replication Rule

If you click on the name of the Replication Rule (Resource column) in the Replication tab, you will be able to view and edit the details of the rule:

View Rule
Figure 127. Replication Rule Details

The View/Edit tab lets you view details or make changes to selected properties of the Rule:

Writer

Indicates which node is set to handle writes for this rule in the local zone. It can be reassigned by selecting the Reassign option, see Writer selection for guidance.

Edit Rule Name

Displays the correct Rule name in gray, this can be edited by entering a new name and selecting the Update button.

Path for this zone

The file path for the replication rule in question and its size.

Zones

The zones that are included for the replication rule.

Priority Zone

You can change the Priority Zone if the rule exists on an even number of Fusion nodes. If there is an odd number of nodes, the Priority Zone selection is only available with an even number of Nodes message will display.

Advanced Options

Various advanced options that can be set for a replication rule. See Advanced Options for guidance.

Remove a Rule

To remove a rule, simply tick the checkbox next to the no longer needed rule(s) and click Remove.

view
Figure 128. Remove a Replication Rule

6.4.3. Path Mapping

As of Fusion version 2.14, it is possible to replicate content between different directories in each Zone.

Example

Path to be replicated

Zone A = /repl/source_path

Destination paths

Zone B = /repl/destination_path_X

Zone C = /repl/destination_path_Y

This can be configured when creating a Replication Rule, please see the Create a rule section for details on how to do this.

6.4.4. Username Translations

From 2.14, per-zone username translation is configurable via the UI. Prior to this release, it was configurable through the application.properties file.
Per-zone username translation allows mappings to be defined based on the name of the originating zone and the username, rather than just the username.

If there is no username translation rule for the requesting zone, the default is to not translate.

Constraints on Username mapping
  • All nodes of the same zone must have the same translation - this happens automatically as any change made in the UI is replicated to all nodes and zones.

  • Mappings only exist as a one-way relationship between two zones, regardless of how many zones are inducted.

  • The source and destination zones must be different. You cannot translate users in the local zone using this tool.

  • Translated users must have appropriate permissions on the destination directories (i.e. write privileges).

  • It is not possible for a username to be matched against more than one rule. If a clash is found, the edit will be rejected.

  • If the translated username defined for the destination zone is not found, the original (i.e. source) username will be used.

  • Once a rule is created, it will take up to a maximum of 20 seconds to become active. If the username translation is not active after this point, please check the Fusion UI logs or contact WANdisco support.

Usernames are not included in consistency checks if username translation is enabled. If any nodes that take part in a consistency check have the Username Translation feature enabled, then inconsistencies in the user field will be ignored. Therefore, when making zones consistent, usernames will remain different for files and directories that are otherwise consistent.
How to set up Username Translations

To set up username translation, see the Username Translation Setup section.

6.4.5. Consistency Check

The Consistency Check tool is used to confirm that the files and metadata that are managed under a replication rule are consistent between zones. Perform consistency checks when you add new data into replication, periodically to monitor the state of replication or as part of a make consistent operation.

Username Translation
If any nodes that take part in a consistency check have the Username Translation feature enabled, then inconsistencies in the "user" field will be ignored.
cc
Figure 129. Consistency Check

A status message appears at the top of the panel that indicates if and when a consistency check has recently been performed on this path.

You can click on Reload results to view the cached results from a previous check, or you can click Trigger check to run a new check.

Path

The path to the replication rule currently being viewed for consistency.

Properties

The system properties for the rule, including the following properties:

  • Length: - byte length of the file (in kilobytes)

  • Is a directory: - distinguishes files from directories (true or false)

  • Owner: - Owning system account

  • Group: - Associated system account group

  • Permissions: - File permissions applied to the element

  • ACLs: - Associated Access Control Lists for the element

Zone columns

Columns will appear for each replicated zone that should contain a copy of the available metadata, as labeled in the Properties field.

Source of truth

From the available zones, you must choose the one that represents the most up-to-date state.

Resolve

Once you have selected from the available zones, click the Resolve button.

cc
Figure 130. replication rule - Preserve extraneous files
Reload results

Click to view the stored results from the previous consistency check. Viewing the results does not require that another check be run but the results may not be up-to-date.

Trigger check

Click to run a new consistency check. This consistency check is non-blocking and is the default type of check that is made through the tool.

Non-blocking Consistency Check
The non-blocking consistency check mechanism (introduced in Fusion 2.11) allows information on consistency state to be determined without blocking other activity while the check is underway. It takes advantage of tracking the state of changes to content under check during execution and produces information for each item checked that covers the states: consistent, not-consistent, potentially inconsistent.
Trigger blocking check

Click to perform a consistency check using the pre-2.11 consistency check mechanism. The new mechanism flags any path that has been changed as "dirty," i.e., cannot be evaluated for consistency. So, there may be environments, with high rates of change, where you need to block further activity during a consistency check to get any useful information about consistency.

Performing a Consistency Check

There are two different ways of doing a Consistency Check:

Consistency Check by Checkbox
  1. Select a Path from the Replication Rules table using the check box column.

    cc
    Figure 131. Consistency Check - Checkbox
  2. The rule-specific options on the top of the panel will no longer be greyed-out. Click on the Consistency Check button.

    cc
    Figure 132. Consistency Check - Check
  3. The selected rule will now be checked in the background.

    cc
    Figure 133. Consistency Check - Check Pending

The results will appear in the Consistency column as either "Unchecked", "Consistent" or "Inconsistent". This result is also a link to the Consistency Check tab.

You can go to to the Consistency Check tab by clicking on the status link in the Status column, or the Path link.

cc
Figure 134. Consistency Check - click link
  1. On the Consistency Check tab, trigger the type of Consistency check you want - see above for the differences between blocking and non-blocking consistency checks.

    cc
    Figure 135. Consistency Check tab - Trigger Check
  2. Based on the results, you may want to run a make consistent operation.

Scheduled Consistency Check

The Consistency Check section under the Settings tab allows you to enable or disable scheduled consistency checks, alter the default check interval and change the consistency check report limit.

See the Consistency Check section for further guidance.

It is also possible to set a different schedule for each specific replication rule using the Advanced Options available when setting up or editing a Replication Rule. See Create a Rule or View/Edit Replication Rule for details.

6.4.6. Make Consistent

The Make Consistent tab provides a tool for resolving file inconsistencies between available zones. The tool provides two different types of operations, based on the option you select from the Make Consistent Type dropdown.

dropdown
Figure 136. Make Consistent Type Dropdown
HCFS Make Consistent

this is an operation on live Hadoop Compatible File System. This method is the most direct for making file systems consistent, although running it will stop writes to the replication rule in the local zone. The block is removed once the operation completes.

Checkpoint Make Consistent

this option uses the fsimage checkpoints created by Hadoop’s admin tool. The use of a snapshot from the namenode ensures that the local filesystem does not get locked during the operation.

Known issue
This option is not available/supported if NameNode Proxy or the Live Ranger / Live Sentry plugins are installed.
Blocking vs Non-blocking Make Consistent

From Fusion 2.12 onwards, make consistent operation defaults to a new non-blocking mechanism that does not block other operations that may be attempted against a location currently being made consistent while that task is in process. This can make the operations less obtrusive, although there are still some situations where the old blocking form of make consistent operation may be required.

By not blocking other activities, during its operation, a Non-blocking Make Consistent can only offer a best-effort attempt to resolve metadata inconsistencies, and may suffer from failures if data that is subject to the make consistent is modified after the sync was initiated.

Limits of Non-blocking Make Consistent
  • Non-blocking make consistent operations are not compatible with checkpoint repairs.

  • Consistency checks cannot be executed in parallel to non-blocking make consistent operations.

  • Non-blocking make consistent operations should not be allowed to run when bypass is enabled.

Make Consistent type HCFS
HCFS breakdown
Figure 137. HCFS make consistent

Run through the following procedure to perform a make consistent operation:

  1. Select the Source of truth from the dropdown. This will flag one of the available zones as most up-to-date / most correct in terms of stored data.

  2. Select from three Resolution types, Recursive, Replace or Preserve (select all that apply). The Resolution description will alter based on your selection.

    Recursive

    If checkbox is ticked, this option will cause the path and all files under it to be made consistent. The default is true, but is ignored if the path represents a file.

Clarification on running without Recursive operation

A make consistent operation that is run without the Recursive option will not work successfully on the contents of any subfolders in the target directory of your replication rule, however it will apply to all contained files and folders, but not the contents of those subfolders.

Example make consistent operation run on /mnt/nfs/dir1 with "no recursive" option.
/mnt/nfs/dir1 - made consistent (no recursive)
/mnt/nfs/dir1/file1 - made consistent
/mnt/nfs/dir1/file2 - made consistent
/mnt/nfs/dir1/subdirectory/file3 - not made consistent
/mnt/nfs/dir1/subdirectory/file4 - not made consistent
Replace

If checkbox is ticked, when the make consistent operation is executed in a zone that is not the source zone, any duplicate files and directories will be overwritten.

Opting not to replace produces inconsistent results when there are pre-existing directories or files. Specifically, if a user attempts to make a path consistent which has files of the same name which exist in both zones, these files will not be affected by the repair if replace is not selected. This will leave the path inconsistent once the repair completes.

The following table shows what to expect when running (and finishing) a Blocking Make Consistent with the Replace checkbox unticked (no attempt to overwrite pre-existing data). To be clear, the outcomes for files or directories are inconsistent.

Table 3. Results from a Blocking Make Consistent with no Replace option
Local Directory Exist Local Files Exist Make Consistent Outcome Remaining

No

No

Make Consistent "complete" (Operation completes as expected)

= 0

No

Yes

Make Consistent remains "incomplete"

= 0

Yes

No

Make Consistent "complete"

> 0

Yes

Yes

Make Consistent "incomplete"

> 0

Preserve

If checkbox is ticked, when the make consistent operation is executed in a zone that is not the source zone, any data that exists in that zone but not the source zone will be retained and not removed.

Non-blocking

When checkbox is ticked, the file system changes that take place after the make consistent operation has been started. This option is ticked by default. You might untick the option if your replicated data environment has high rates of change, where you need to block HDFS client operations during a consistency check to get any useful information about consistency.

Make consistent type Checkpoint
HCFS Make Consistent
Figure 138. Checkpoint type Make Consistent operation

Known issue
This option is not available/supported if NameNode Proxy or the Live Ranger / Live Sentry plugins are installed.
Example Make Consistent operation
  1. Compare the inconsistent states of the consistency check result. You need to identify which version of the file/metadata is correct/most up-to-date, then set the zone on which it is located as the Source of truth.

    NameNodeSettings
    Figure 139. Bulk Make Consistent
  2. Once selected, click Resolve.

    NameNodeSettings
    Figure 140. Resolve
  3. Please select any rules that you wish to apply to the make consistent operation. You can see an explanation for each action in the section Make Consistent type HCFS.

    NameNodeSettings
    Figure 141. Make Consistent handling rules
  4. A confirmation message will display the status of the make consistent operation. You should see "Fix requested" appear in a green background in the Source of truth box. This indicates that the fix has been submitted. You will need to rerun the consistency check to confirm that the issue is fixed.

    NameNodeSettings
    Figure 142. Fix requested
Bulk make consistent

When a large number of inconsistencies are found, you don’t need to manually set each make consistent operation, instead you can use the Bulk make consistent tool to set the rules for a collective operation that applies to multiple paths.

  1. Select Bulk resolve inconsistencies from the dropdown, this will reveal the make consistent tool screen. The make consistent tool works in the same way as the regular make consistent operation. Read the start of this section for more information on the Make Consistent tool options.

    NameNodeSettings
    Figure 143. Bulk Make Consistent
  2. Once you start an operation, you will get a confirmation request. Click Confirm to continue.

    NameNodeSettings
    Figure 144. Bulk Make Consistent confirmation
  3. Next, you will receive a confirmation message that the request has been submitted successfully.

    NameNodeSettings
    Figure 145. Request submitted

    Rerun the Consistency Check to confirm that all issues have now been resolved.

Running initial make consistent operations

If you have a large directory you can parallelize the initial make consistent operation using the Fusion API. This can be accomplished on a single file or a whole directory. Choosing a directory will push all files from the source to the target regardless of existence at the target.

Consider the following directory structure for a fusion replication rule /home

/home
/home/fileA
/home/fileB
/home/userDir1
/home/userDir2
/home/userDir3

We could run a bulk resolve in the UI against the /home directory, however, to provide parallelism of the make consistent operations we can use the Fusion API to issue make consistent operations against each directory and the individual files in the /home directory.

REST API Call
"FUSION_NODE:PORT/fusion/fs/repair?path=SYSTEMPATH&recursive=true&src=ZONENAME"
Example - Multiple API Calls using curl
curl -X PUT "FUSION_NODE:8082/fusion/fs/repair?path=/home/userDir1&recursive=true&src=LocalFS"
curl -X PUT "FUSION_NODE:8082/fusion/fs/repair?path=/home/userDir2&recursive=true&src=LocalFS"
curl -X PUT "FUSION_NODE:8082/fusion/fs/repair?path=/home/userDir3&recursive=true&src=LocalFS"
curl -X PUT "FUSION_NODE:8082/fusion/fs/repair?path=/home/fileA&recursive=false&src=LocalFS"
curl -X PUT "FUSION_NODE:8082/fusion/fs/repair?path=/home/fileB&recursive=false&src=LocalFS"

This will spawn simultaneous make consistent operations, increasing the performance of the initial synchronization. This is especially helpful when you have small file sizes to better saturate the network.

For files, the recursive parameter is ignored
You can use the file transfers view in the Fusion UI on the OpenStack-replicating node to monitor the incoming files.

Cancelling a make consistent operation

In the event that a make consistent operation gets stuck, or for some reason needs to be cancelled, this can be actioned from the Make Consistent tab. Ongoing make consistent operations are noted with a label on the Replication Rule screen.

Make consistent cancellation
Figure 146. Fusion make consistent operation underway label

Cancel an ongoing make consistent operation by going to the Make Consistent History panel on the Make Consistent tab and clicking the Cancel Make Consistent button.

NameNodeSettings
Figure 147. Fusion make consistent cancellation
Started

Identifies the rule path under make consistent operation. Click the Details button for more information about the state of the process.

Local Status

Will show as Completed or Making Consistent.

Initiating Node

Shows the node from which the make consistent proposal was created.

Started

Time-stamp for the start of the make consistent.

Elapsed

Amount of time that the make consistent operation has been underway.

Options

Lists the make consistent options selected for the operation. For an explanation of their meaning, see Make Consistent Type HCFS.

Cancel Make Consistent Operation

If an operation is still underway, there is a Cancel button available that lets you halt it.

6.4.7. File Transfers

The File Transfer panel shows the movement of data coming into the zone. The current transfer speed is displayed.

cc
Figure 148. File Transfer

The File Transfer list can be filtered or sorted by column.

6.4.8. Bandwidth management

Enterprise License only
The Bandwidth Management tools are only enabled on clusters that are running on an Enterprise license. See the Deployment Checklist for details about License Types.

The bandwidth management tools provide two additional areas of functionality to support Enterprise deployments.

  • Limit the rate of outgoing traffic to each other zone.

  • Limit the rate of incoming traffic from each other zone.

Any applicable bandwidth limits are replicated across your nodes and applied on a per-zone basis.

The Fusion Nodes screen will display current incoming traffic for the local zone. You will need to log in to the Fusion UI on a node within each Zone to see all incoming traffic levels.

Setting up bandwidth limits

Use this procedure to set up bandwidth limits between your zones.

On the Nodes page in the Fusion UI, click on the Set bandwidth limits button that will be present for the local zone.

Bandwith
Figure 149. Set bandwidth limits

The Maximum bandwidth dialog will open. For each remote zone you can set maximum Outgoing and Incoming values, either by typing in your value or using the arrows in the box. Entered values are in Mibit (Megabits).
The default value is 0 which indicates unlimited bandwidth.

Bandwith
Figure 150. Maximum bandwidth entry dialog
Outgoing

The provided value will be used as the bandwidth limit for data being sent from the local zone.

Incoming

The provided value will be used as the bandwidth limit for data being received into the local zone. Note that any Incoming value applied will also change the remote zone(s) Outgoing limit for data being sent to this local zone.

When you have set your bandwidth values, click Update to apply these settings to your deployment.

6.4.9. Reporting

The following section details the reporting tools that WANdisco Fusion currently provides.

Consistency Check

The consistency check mechanism lets you verify that replicated HDFS data is consistent between sites. Read about Handling file inconsistencies.

Consistency Checks through WANdisco Fusion UI
Username Translation
If any nodes that take part in a consistency check have the Username Translation feature enabled, then inconsistencies in the "user" field will be ignored.
NameNodeSettings
Figure 151. Replication Rules table - indicates if inconsistencies are detected.
Consistency
Consistency Status

A status which links to the consistency check report. It can report Check Pending, Inconsistent, Consistent or Unknown.

Last Check

Shows the time and date of the check that produced the current status. By default, Consistency checks are not automatically run.

Next Check

Shows the time and date of the next automatically scheduled Consistency Check. Remember, you don’t need to wait for this automatic check, you can trigger a consistency check at any time through the Consistency Check tool.

Click on the Consistency/Inconsistency link to get more information about the consistency check results for a selected path.

Read more about Consistency Check tool.

File Transfer Report

As a file is being pulled into the local zone, the transfer is recorded in the WANdisco Fusion server and can be monitored for progress.

Use the REST API filter by the replicated path and sort by ascending or descending "complete time" or "start time":

GET /fusion/fs/transfers?path=[virtualPath]&sortField=[completeTime*|startTime]&order=[DESC*|ASC]&showAll=[false*|true]

In the above call defaults are marked as *.

File transfer Report Output

Example output showing an in-progress and completed transfer:

<fileTransfers>
    <fileTransfer>
        <startTime>1426020372314</startTime>
        <elapsedTime>4235</elapsedTime>
        <completeTime>1426020372434</completeTime>
        <username>wandisco</username>
        <familyRepresentativeId>
            <nodeId>f5255a0b-bcfc-40c0-b2a7-64546f571f2a</nodeId>
            <dsmId>93452fe3-c755-11e4-911e-5254001ba4b1</dsmId>
        </familyRepresentativeId>
        <file>/tmp/repl/isoDEF._COPYING_<;/file>
        <remoteFs>hdfs://vmhost5-vm4.frem.wandisco.com:8020</remoteFs>
        <origin>dc1<;/origin>
        <size>4148166656</size>
        <remaining>4014477312</remaining>
        <bytesSec>3.3422336E7</bytesSec>
        <percentRemaining>96.77714626516683</percentRemaining>
        <state>in progress</state>
    </fileTransfer>

    <fileTransfer>
        <startTime>1426019512082</startTime>
        <elapsedTime>291678</elapsedTime>
        <completeTime>1426019803760</completeTime>
        <username>wandisco</username>
        <familyRepresentativeId>
            <nodeId>f5255a0b-bcfc-40c0-b2a7-64546f571f2a</nodeId>
            <dsmId>93452fe3-c755-11e4-911e-5254001ba4b1</dsmId>
        </familyRepresentativeId>
        <file>/tmp/repl/isoABC</file>
        <remoteFs>hdfs://vmhost5-vm4.frem.wandisco.com:8020</remoteFs>
        <origin>dc1</origin>
        <size>4148166656</size>
        <remaining>0</remaining>
        <bytesSec>1.4221733E7</bytesSec>
        <percentRemaining>0.0</percentRemaining>
        <state>complete</state>
    </fileTransfer>
</fileTransfers>
Output key with data type
Username

System user performing the transfer. (String)

File name

Name of the file being transferred. (String)

Remote FS

The file of the originating node. (URI)

Origin

The file’s originating Zone. (String)

Size

The cumulative size of data transferred. (Long)

Appends

The number of appends that have been made to the file being transferred. (Long)

AppendSize

The size of the latest append.

Remaining

Remaining bytes still to be transferred for the latest append. (Long)

Percent remaining

Percentage of the file still to be transferred. (Double)

Bytes/Sec

The current rate of data transfer, i.e. Amount of file downloaded so far / elapsed download time. (Long)

State

One of "in progress", "incomplete", "completed", "appending", "append complete", "deleted" or "failed". (TransferState)
In progress: means we are performing an initial pull of the file.
Appending: means data is currently being pulled and appended to the local file.
Append completed: means all available data has been pulled and appended to the local file, although more data could be requested later.

Note: files can be renamed, moved or deleted while we pull the data, in which case the state will become "incomplete".
When the remote file is closed and all of its data has been pulled, the state will then change to "Complete".
If a file is deleted while we are trying to pull the end state will be "deleted".
If the transfer fails the state will be "failed".

Start Time

The time when the transfer started. (Long)

Elapsed Time

Time that has so far elapsed during the transfer. Once the transfer completes it is then a measure of the time between starting the transfer and completing. (Long)

Complete Time

During the transfer this is an estimate for the complete time based on rate of through-put so far. Once the transfer completes this will be the actual time at completion. (Long)

Delete Time

If the file is deleted then this is the time the file was deleted from the underlying filesystem. (Long)

6.4.10. Record retention

Records are not persisted and are cleared up on a restart. The log records are truncated to stop an unbounded use of memory, and the current implementation is as follows:

For each state machine, if there are more than 1,000 entries in its list of transfers we remove the oldest transfers ,sorted by complete time, which are in a terminal state ("completed", "failed" or "deleted") until the size of the list is equal to 1,000. The check on the number of records in the list is performed every hour.

6.4.11. Renaming files and directories

Files and directories under replication can be renamed. However if one or more zones is encrypted, and the others are not, then the rename may fail. See the Hadoop documentation for more information.

This rename failure will happen in the following circumstances:

  • The file/directory in the originating cluster is not in an encrypted zone, or the rename means the file/directory remains within the same encrypted zone, but, the rename will result in the file/directory moving into or out of an encrypted zone in the target cluster.

    • Result: The rename will complete on the source cluster but will fail in target clusters.

  • When the rename of a file/directory in the originating cluster will result in a move out of, into or between encrypted zones, but, in the target cluster there will be no movement in relation to encrypted zones.

    • Result: The rename will fail on the source cluster but will complete in target clusters.

Example

  • The path /repl1/a is defined in a replication rule and it contains the file or directory /repl1/a/b/c.

  • In Zone A, there is no encrypted zone. In Zone B, there is an encrypted zone of /repl1/a/b.

In this set up, if the command hdfs dfs -mv /repl1/a/b/c /repl1/a/X/c is run on Zone A, then the rename will work on Zone A but it will fail on Zone B. This is because c is being moved from within the encrypted zone /repl1/a/b to outside of it in /repl1/a/X.

6.5. Administration

6.5.1. Manual Bypass

Manual bypass allows clients to bypass WANdisco Fusion. It can be used for WANdisco Fusion maintenance and troubleshooting.

When manual bypass is enabled, consistency check and make consistent can still continue in both directions. Replication can also continue from the remote zone to the zone with the bypass in place.

The warning 'This zone is not actively replicating' is displayed on the Fusion dashboard when manual bypass is enabled.

Enable Client Bypass
Figure 152. Manual Bypass enabled
Connecting to Fusion

Client access to Fusion is mostly defined by URI selection. This is specified during installation, or via the Settings tab on the Fusion UI of an established deployment. It can also be controlled manually by amending the cluster core-site directly, though your cluster manager (CM, Ambari, etc)

URI Selection

There are two parameters that are modified according to your needs. You can modify the default HDFS "implementation" (fs.hdfs.imp) to point at Fusion instead of the default Hadoop HDFS client, or we add a whole new "fusion". Note the fourth option, in the table below; you can run both URIs in parallel.

URI core-site parameter points to Fusion class:

hdfs:// with HDFS

fs.hdfs.impl

com.wandisco.fs.client.FusionHdfs

fusion:// with HCFS

fs.fusion.impl

com.wandisco.fs.client.FusionHcfs

fusion:// with HDFS

fs.fusion.impl

com.wandisco.fs.client.FusionHdfs

fusion:// & hdfs:// with HDFS

fs.hdfs.impl & fs.fusion.impl

both point to com.wandisco.fs.client.FusionHdfs

You can create your own implementation by defining a new .impl parameter and pointing it at a class that handles the filesystem commands of your choice. E.g., fs.fusion.impl could be renamed (or exist in conjunction with) fs.secret.impl, which would mean any filesystem commands to path prefixed with secret:// get handled by the Fusion class as well.
Bypass Option

With the HDFS URI, Fusion sits in the write path for all cluster filesystem edits, so a complete Fusion outage would prevent any cluster write operations. The bypass option allows clients to write directly to the local cluster, preventing a block on cluster activity. The bypass can be enabled from the UI or through the following manual process.

<property>
	<name>fusion.client.can.bypass</name>
	<value>true</value>
</property>

By default, when the parameter is missing, it’s set to false. With it set to true, if all the Fusion servers are down in a zone then the client writes directly to the underlyingFS. If you perform any client operations via command line you should see the following additional output:

Bypassing enabled with ResponseWaitTime(secs): X, RetryInterval(secs): Y.

You may also see the following additional output after it bypasses to the underlying:

Bypassing triggered for Y seconds.

The other parameter to adjust the response time and retry interval for long-running clients:

<property>
	<name>fusion.client.bypass.retry.interval.secs</name>
	<value>integer number representing seconds; default is 60</value>
</property>
fusion.client.bypass.retry.interval.secs

is the number of seconds for which the client will stay "bypassed" before trying Fusion again.

Hiding the Fusion servers and Manual Fast Bypass

The bypass procedure is not just for handling node failure, there may be occasions that you wish to hide Fusion from the clients so that bypass happens while Fusion servers are up and running, during maintenance, application troubleshooting, or performing tests commands with and without Fusion, etc.

The reason for doing this are:

  • it enables consistency check and make consistent to continue operating, in both directions.

  • it allows replication to continue from the remote zone to the zone with the bypass in place

  • it prevents stopped nodes from accidentally sidelining if down too long

  • it allows individual clients to continue contacting Fusion server for testing / troubleshooting (if they’re then configured without repl_exchange_dir)

A Manual Fast Bypass flag available as part of the replicated exchange directory feature. If your clusters is configured with the replicated_exchange_dir then adding a subdirectory called "bypass" to the repl_exchange_dir tells your Fusion Clients to treat all folders as non-repl, and to therefore not contact the Fusion server at all. This is, in effect, a quick way to hide the Fusion servers. Clients will continue to bypass Fusion until the "bypass" flag is deleted.

bypass parameter
<property>
<name>fs.fusion.client.can.bypass</name>
<value>true</value>
</property>
  • By default the parameter is missing, so is therefore "false".

  • When set to true, if all the Fusion servers are down in a zone then the client writes directly to the underlyingFS. If you perform any client operations via command line you should see the following additional output:

Bypassing enabled with ResponseWaitTime(secs): X, RetryInterval(secs): Y.

You may also see the following additional output after it bypasses to the underlying:

Bypassing triggered for Y seconds.
Bypass utility script

Using the bypass utility script, bypass paths can be created under the configured replicated exchange directory. For more information on using this script see the Knowledge base article Fusion Bypass Utility Script.

6.5.2. Starting up

To start WANdisco Fusion UI:

  1. Open a terminal window on the server and log in with suitable file permissions.

  2. Run the fusion-ui-server service from the /etc/init.d directory:

    rwxrwxrwx  1 root root    47 Apr 10 16:05 fusion-ui-server -> /opt/wandisco/fusion-ui-server/bin/fusion-ui-server
  3. Run the script with the start command:

    # ./fusion-ui-server start
    
    Starting fusion-ui-server:.                         [  OK  ]

    WANdisco Fusion starts. Read more about the fusion-ui-server init.d script.

  4. Also you can invoke the service directly. e.g.

    # service fusion-ui-server stop/start

6.5.3. Shutting down

To shut down:

  1. Open a terminal window on the server and log in with suitable file permissions.

  2. Run the WANdisco Fusion UI service, located in the init.d directory:

    rwxrwxrwx  1 root root    47 Dec 10 16:05 fusion-ui-server -> /opt/wandisco/fusion-ui-server/bin/fusion-ui-server
  3. Run the stop script:

    # ./fusion-ui-server stop
    stopping fusion-ui-server:                                   [  OK  ]

    The process shuts down.

Shutdowns take some time

The shutdown script attempts to stop processes in order before completing, as a result you may find that (from WANdisco Fusion 2.1.3) shutdowns may take up to a minute to complete.

6.5.4. init.d management script

The start-up script for persistent running of WANdisco Fusion is in the /etc/init.d directory. Run the script with the help command to list the available commands:

# service fusion-ui-server help
  usage: ./fusion-ui-server (start|stop|restart|force-reload|status|version)

start Start Fusion services
stop Stop Fusion services
restart Restart Fusion services
force-reload Restart Fusion services
status Show the status of Fusion services
version Show the version of Fusion

Check the running status (with current process ID):

# service fusion-ui-server status
Checking delegate:not running                              [  OK  ]
Checking ui:running with PID 17579                         [  OK  ]

Check the version:

# service fusion-ui-server  version
1.0.0-83

6.5.5. NameNode Proxy service management

If you need to restart the NameNode Proxy service then go to the Cluster Manager and restart the service there.

If running the stop command, you must ensure that Fall-through is in place and activated. If there is no form of failsafe in place, client requests will not reach the NameNode.

6.5.6. Managing cluster restarts

WANdisco Fusion’s replication system is deeply tied to the cluster’s file system (HDFS). If HDFS is shut down, the WANdisco Fusion server will no longer be able to write to HDFS, stopping replication even if the cluster is brought back up.

To avoid replication problems:

  • Where possible, avoid doing a full shutdown. Instead, restart services to trigger a rolling restart of datanodes.

  • If a full shutdown is done, you should do a rolling restart off all WANdisco Fusion nodes in the corresponding zone. A rolling restart ensures that you will keep the existing quorum.

6.5.7. Managing services through the WANdisco Fusion UI

Providing that the UI service is running, you can stop and start WANdisco Fusion through the Nodes tab.

Dashboard
Figure 153. Stop Node via Fusion Node tab

The Fusion Nodes table shows the:

Name

The node name. Labels in this column indicate which node is the local node or give links to the remote nodes.

Jump to node/Local Node

Button indicates the Local Node or provides a link to other nodes. The address to jump between nodes can be changed on the Settings page of the relevant node.

Fusion Status

The status of the node.

Last updated

The time and date of the last change on the node.

IP Address:Port

The IP address and Fusion port (default is 6444).

From the top bar of the table you can induct a new node as well as stop and start nodes.

6.5.8. WANdisco Fusion UI login

The UI for managing WANdisco Fusion can be accessed through a browser, providing you have network access and the port that the UI is listening on is not blocked.

http://<url-for-the-server>:<UI port>

e.g.

http://wdfusion-static-0.dev.organisation.com:8083/ui/

You should not need to add the /ui/ at the end, you should be redirected there automatically.

dashboard
Figure 154. Dashboard

Log in using your Hadoop platform’s manager credentials or applicable LDAP/AD credentials

Login credentials

Currently you need to use the same username and password that are required for your platform manager, e.g. Cloudera Manager or Ambari. In a future release we will separate WANdisco Fusion UI from the manager and use a new set of credentials.

LDAP/Active Directory and WANdisco Fusion login

If your Cloudera-based cluster uses LDAP/Active Directory to handle authentication then please note that a user that is added to an LDAP group will not automatically be assigned the corresponding Administrator role in the internal Cloudera Manager database. A new user in LDAP that is assigned an Admin role will, by default, not be able to log in to WANdisco Fusion. To be allowed to log in, they must first be changed to an administrator role type from within Cloudera Manager.

Running without LDAP-based Authn/Authz (CDH)

CDH holds its own database of Cloudera Manager users, and has authz roles for them (RO, Full Admin etc.) We only allow Fusion UI logins for manager users who have full admin privileges.

Note that the internal database is NOT synchronized with the external auth providers, e.g., there may be an LDAP group that CDH regards as full admin, but the role is NOT entered into the internal database, so without manual intervention, these users will be denied access to the Fusion UI.

The above functionality appears to be session-based, in that, if one of these users logs in to the manager and views their profile, it could indicate that they have full admin privileges. The simple workaround is the use Fusion’s LDAP integration.

For more information about setting up the LDAP/AD sync, see LDAP/ Active Directory.

Change login credentials

The procedure to change the user/password credentials differs depending on whether you are using a Fusion with a manager or without a manager (e.g. cloud deployments), in the zone.

With a manager

When using a manager, you need to use your manager’s username and password. To make changes to authentication you need to handle this through your manager. For information about how to do this, you will need to review your manager’s documentation.

UI authentication of the manager

If you are using a zone with a manager use the following steps to change the UI’s authentication for your manager, which allows it to inspect settings and cluster status and update configuration.

  1. Generate a new hash for your password using:

    /opt/wandisco/fusion/tools/bin/encrypt-password.sh
  2. On each Fusion server, edit /opt/wandisco/fusion-ui-server/properties/ui.properties and update the following properties with the new credentials:

    manager.username
    manager.encrypted.password
  3. Restart Fusion UI server, see init.d management script.

    service fusion-ui-server restart
  4. Repeat procedure for all WANdisco Fusion server nodes in the zone. Note that you can reuse the hash generated in step 1, you do not need to create a new one for each password.

Without a manager

This section covers the use of login credentials that are not controlled by a cluster manager.

Generate a new password

If you need to reset a lost password, or if you don’t have a Hadoop manager in the zone, for example if you are using a Cloud zone, use the following steps to set, change or reset lost login credentials:

  1. Generate a default admin user.password using the following command.

    manager user (access to the UI but not LDAP or super user settings)
    cd /opt/wandisco/fusion-ui-server
    java -cp fusion-ui-server.jar com.wandisco.fusionui.authn.ResetPasswordRunner -f /opt/wandisco/fusion-ui-server/properties/ui.properties

Confirm your new password. This will automatically update the user.password property in the ui.properties file. This file also may contain the properties super.user.username, for Fusion Administrator and manager.username, for the default manager user, which are used for storing the account usernames, and may be updated with a manual edit, if required.

  1. Restart Fusion UI server, see init.d management script.

    service fusion-ui-server restart
  2. Repeat procedure for all WANdisco Fusion server nodes in the zone.

Generate a new password for the Fusion Administrator

This function is applicable for deployments where a Fusion Administrator has been set up, for which the deployment must be running with a cluster manager.

Follow the instructions provided, above with the following command, used instead:

Fusion Administrator password change
cd /opt/wandisco/fusion-ui-server
java -cp fusion-ui-server.jar  com.wandisco.fusionui.authn.ResetPasswordRunner -s -f /opt/wandisco/fusion-ui-server/properties/ui.properties

6.5.9. Authentication misalignment

There are four possible scenarios concerning how LDAP authentication can align and potentially misalign with the internal CM database:

User has full access in CM, denied access in WANdisco Fusion UI
  • User is in the Full Administrator group in LDAP

  • User is left as the default read-only in the internal Cloudera Manager database

User has full access in CM, full access in WANdisco Fusion UI
  • User is in the Full Administrator group in LDAP

  • User is changed Full Administrator in the internal Cloudera Manager database

User has read-only access in CM, denied access to WANdisco Fusion UI
  • User is removed from the Full Administrator group in LDAP and added to the read-only group

  • User is left as the default read-only in the internal Cloudera Manager database

User has read-only access to CM, Full access to WANdisco Fusion UI
  • User is removed from the Full Administrator group in LDAP and added to the read-only group

  • User is set as Full Administrator in the internal Cloudera Manager database + Clearly this scenario represents a serious access control violation, administrators must audit WANdisco Fusion users in Cloudera Manager.

6.5.10. Set up email notifications

Email notifications can be set up in the Fusion UI for monitoring purposes, there are multiple scenarios in which emails can be sent.

An SMTP server is also required to enable this feature, see the Email Notifications section for guidance on setting up this feature.

6.6. Maintenance

6.6.1. Writer selection

A writer is the term for a node that is responsible for all coordinated write operations on a given replicated path. There is one writer assigned for each path in a Zone.

From Fusion version 2.14 onwards, the writer for a replicated path can be changed within the Fusion UI.

This is used for the purposes of load balancing, and manual recovery from a node failure.

The Fusion API must not be used to invoke this feature. The Fusion UI should always be used, as it has additional safeguards that help prevent the re-execution of events.

There are two methods to which writers can be reassigned, either by abdicating or revoking a node’s current assignment. This can be performed at the Node or Rule level.

You must be on the Fusion UI of a node in the same Zone to perform any writer changes.
Node level

This section will detail how to make writer changes at the node level. The node page of a given node can be accessed via the Nodes page in the Fusion UI.

Once on the Nodes page, select the node name in blue text to access its details.

wdf2.14 nodes page
Figure 155. Nodes page
Abdicate

When on the node page for a node which is healthy, the user can give away the writer status for a chosen number of rules, to another node in the zone.

wdf2.14 node level abdicate writer
Figure 156. Node level - abdicate

When selecting to Reassign, the user will then define how many replicated rules are to be reassigned, as well as selecting the new writer node.

wdf2.14 node level reassign options
Figure 157. Node level - reassign options

Once clicking on Confirm, the replicated paths will then be reassigned to the new writer.

wdf2.14 node level abdicating
Figure 158. Node level - abdicating
Revoke

When on the node page for a node which is unhealthy, the user is shown a button (Reassign), allowing them to revoke the writer status for all of the rules to which that node is the writer. This is a destructive action as it informs the rest of the nodes that this node is no longer operable, and therefore alters the state of the ecosystem by remapping the replicated path(s) to another node.

wdf2.14 node level revoke writer
Figure 159. Node level - revoke writer

Once selecting the writer to be assigned the replicated rules, and clicking Confirm, the replicated paths will be revoked from the unhealthy node and reassigned to the new writer.

wdf2.14 node level revoking
Figure 160. Node level - revoking
Rule level

This section will detail how to make writer changes at the replicated rule level. The page of a given replicated rule can be accessed via the Replication page in the Fusion UI.

Once on the Replication page, select the replicated rule in blue text (Resource column) to access its details.

wdf2.14 replication page
Figure 161. Replication page
Abdicate

When on the page for a replicated rule where the writer node is in a healthy state, the user can select another node to become the writer for that rule.

wdf2.14 rule level abdicate writer
Figure 162. Rule level - abdicate

When selecting to Reassign, the user will select the new writer node.

wdf2.14 rule level reassign options
Figure 163. Rule level - reassign options

Once clicking on Confirm, the replicated paths will then be reassigned to the new writer.

wdf2.14 rule level abdicating
Figure 164. Rule level - abdicating
Revoke

When on the page for a replicated rule where the writer node is in an unhealthy state, the user will be offered a link to view the node where they can revoke writer status for all rules assigned to the unhealthy node.

wdf2.14 rule level revoke writer
Figure 165. Rule level - revoke writer

Once selecting Jump to node, the user will be redirected to the node page of the unhealthy node. Once selecting to Reassign, the user will be presented with the same options as detailed in the Node level - Revoke section.

wdf2.14 node level revoke writer
Figure 166. Node level - revoke writer

Once selecting the writer to be assigned the replicated rules, and clicking Confirm, the replicated paths will be revoked from the unhealthy node and reassigned to the new writer.

wdf2.14 node level revoking
Figure 167. Node level - revoking

6.6.2. Tunables

Fine-tuning Replication

WANdisco’s patented replication engine, DConE, can be configured for different use cases, balancing between performance and resource costs. The following section looks at a number of tunable properties that can be used to optimize WANdisco Fusion for your individual deployment.

Increasing thread limit

WANdisco Fusion processes agreements using a set number of threads, 20 by default, which offers a good balance between performance and system demands.

It is possible, in cases where there are many Copy agreements arriving at the same time, that all available threads become occupied by the Copy commands. This will block the processing of any further agreements.

You can set WANdisco Fusion to reserve more threads, to protect against this type of bottleneck situation:

Increase executor.threads property
  1. Make a backup copy of WANdisco Fusion’s applications config file /etc/wandisco/fusion/server/application.properties, then open the original in your preferred text editor.

  2. Modify the property executor.threads.

    Property Description Permitted Values Default Checked at…​

    executor.threads

    The number of threads executing agreements in parallel.

    1-Integer.MAX_VALUE

    250

    Startup

    WANdisco Fusion Server snippet

    Don’t go alone
    Any upward adjustment will clearly increase the resourcing costs. Before you make any changes to replication system properties, you should open up discussions with WANdisco’s support team. Applying incorrect or inappropriate settings to the replication system may result in hard to diagnose problems.

  3. Save your edited application.properties file, then restart WANdisco Fusion.

6.6.3. System Status on the dashboard

The WANdisco Fusion UI dashboard provides a view of WANdisco Fusion’s status. From the Cluster Graph you can identify which data centers are experiencing problems, track replication between data centers or monitor the usage of system resources.

Dashboard
Figure 168. Dashboard

UI Dashboard will indicate if there are problems with WANdisco Fusion on your cluster.

Environment

The environment icon identifies the type of file system being replicated. E.g.

  • Hadoop

  • Cloud

  • Local File System

Zone

The name of the zone that you are viewing.

Node

The name of the node that you are viewing.

Operations

Number of pending operations.

License Status

Shows the type of license in use. Evaluation, Production or Unlimited.

Data volume

Only shown if you are using an Evaluation license. This bar graph displays the volume of used data transfer, as a percentage value. The actual volume of replicated data can be viewed further down on the Dashboard.

Activity Graphs
Activity in the last hour

This graph shows the activity which occurred in the last hour.

Fusion03
Figure 169. License Limit
License Limit: Volume of Replicated Data

On nodes that have data limits on their product license (an Evaluation license), there’s a graph that displays the volume of replicated data, as a percentage of the license limit.

Fusion03
Figure 170. License Limit
Fusion Database Partition Disk Usage

This graph measures the percentage of available storage in the partition that hosts the WANdisco Fusion installation.

F2
Figure 171. Fusion Database Partition Disk Usage
System CPU Load

This graph tracks the current percentage load on the cluster’s processors.

F3
Figure 172. System CPU Load
Graph Settings

The graphs that are displayed on the WANdisco Fusion dashboard can be modified so that they use different thresholds for their "Warning" and "Critical" levels. By default, warn triggers at 80% usage and critical triggers at 90% or 95%.

Warning

At the warn level, the need for administrator intervention is likely, although the state should have no current impact on operation. On a breach, there is the option for WANdisco Fusion to send out an alerting email, providing that you have configured the email notification system. See Set up email notifications.

Crtical

At the critical level, the need for administrator intervention may be urgent, especially if the breach concerns partition usage where reaching 100% will cause the system to fail and potentially result in data corruption. On a breach, there is the option for WANdisco Fusion to send out an alerting email, providing that you have configured the email notification system. See Set up email notifications.

CPU Graph clarification

We display CPU load averages. Low values indicate that the system’s processor(s) have unused capacity. Above the warning threshold (80% by default) available capacity starts to run out. Note that the number that drives the graph is between 0 and 1, and so already takes multi-core systems into consideration.

6.7. Server Logs Settings

The WANdisco Fusion logs that we display in the WANdisco Fusion UI are configured by properties in the ui.properties file.

logs
Figure 173. Server logging

6.7.1. WANdisco Fusion UI Logs viewer

Using WANdisco Fusion UI’s log viewer (View Logs):

  1. Log in to the WANdisco Fusion UI and click on the Nodes tab button. Then click on the Node on which you wish to view logs.

    logs
    Figure 174. Log viewer 1
  2. Click on the View Logs link, in the Local WANdisco Fusion Server table:

    logs
    Figure 175. Log viewer 2
  3. The View Logs screen lets you select from either WANdisco Fusion or UI Server logs.

    logs
    Figure 176. Log viewer 3

6.7.2. Changing the timezone

Fusion IHC and Server

To alter the timezone the PatternLayout property needs to be overwritten.

<PatternLayout pattern="%d{ISO8601} {UTC} %p %c - %t:[%m]%n"/>

{UTC} can be replaced with, for example {GMT} or {ITC+1:30}. If offsetting from a timezone, + or - can be used, hours must be between 0 and 23, and minutes must be between 00 and 59.

This property is located in several xml files. For an example set up these are listed below, but the exact paths may differ for your set up:

  • /etc/wandisco/fusion/server/log4j2.xml

  • /etc/wandisco/fusion/ihc/server/hdp-2.6.0/log4j2.xml

After updating all the relevant files, the Fusion Server and Fusion IHC Server will take up the new changes after 5 minutes (default).

Fusion UI Server

Logs use UTC timezone by default but this can be manually altered through log4j configuration if required.

To alter the timezone the xxx.layout.ConversionPattern property needs to be overwritten.

log4j.appender.stdout.layout.ConversionPattern=%d{ISO8601}{UTC} %p %c - %t:[%m]%n

{UTC} can be replaced with, for example {GMT} or {ITC+1:30}. If offsetting from a timezone, + or - can be used, hours must be between 0 and 23, and minutes must be between 00 and 59.

This property is located in an xml file. For an example set up this is listed below, but the exact path may differ for your set up:

  • /opt/wandisco/fusion-ui-server/lib/fusion_ui_log4j.xml

After updating the relevant file, the Fusion UI Server will need to be restarted for the changes to take effect.

6.7.3. Logging at startup

At startup the default log location is /dev/null. If there’s a problem before log4j has initialised this will result in important logs getting lost. You can set the log location to a filespace that preserve early logging.

Edit fusion_env.sh adding paths to the following properties:

SERVER_LOG_OUT_FILE

Path for WANdisco Fusion server log output

IHC_LOG_OUT_FILE

Path for IHC server log output

More about logging
For more information about WANdisco Fusion’s logging, see Troubleshooting - Read logs.

6.8. Security

6.8.1. Secure Socket Layer (SSL) encryption

Secure Socket Layer (SSL) encryption can be used to secure Fusion traffic along a number of different network paths. These are:

  • Replication traffic between Fusion Servers and Fusion IHC Servers (i.e. between Zones).

  • Communication between Web Browser and Fusion UI (HTTPS).

  • Client to Fusion Server encryption.

This guide will cover SSL encryption of the paths listed above using the Fusion UI, as well as information on Keystores/Truststores.

The following default ports will change when SSL is enabled
  • Fusion Server HTTP API moves from 8082 to 8084.

  • Fusion IHC Server HTTP API moves from 9001 to 8001.

  • Fusion UI moves from 8083 to 8443.

Check your Java
Pay attention to the version of Java running on your platform. Some versions have bugs with memory leaks or crippled performance when SSL is enabled. For an example, see this link.
Enable SSL for the WANdisco Fusion Server / IHC Server
If configuring Fusion for the first time, the Update button can be selected if not enabling SSL at this time. If you are enabling SSL, follow the guidance in this section.

The following procedures are used for setting up SSL encryption for the WANdisco Fusion server and IHC server.

Before you enable use of HTTPS, ensure that all WANdisco Fusion nodes/zones have been installed and configured, although it is not necessary to have inducted the nodes.

The procedure must be followed for each WANdisco Fusion server in your replicated ecosystem, in turn.

Enable HTTPS on all nodes
If you don’t enable HTTPS on some nodes, some information (e.g. graph data) will not be displayed.
  1. Create a Key Store file using keytool, then save the file to a location on each node where the WANdisco Fusion server can read it.

  2. Log in to WANdisco Fusion UI, click on the Settings tab.

  3. Select to display the Fusion Connection section.

    wdf2.14 settings fusionconnection01
    Figure 177. Settings - Fusion Connection
    "Only HTTPS" or "HTTP and HTTPS"

    Once selected, the additional properties will be listed below:

    Fusion HTTPS Server Port

    The TCP port that will be used for the SSL traffic (default 8084).

    Key Store Path

    Absolute filesystem path to the keystore.
    Example location
    /opt/wandisco/ssl/keystore.ks

    Key Store Password

    Password for the keystore.

    Key Alias

    The Alias of the private key.
    Example
    wandisco

    Key Password

    Password for the Key.

    TrustStore Path

    Absolute filesystem path to the truststore.
    Example location
    /usr/java/jdk1.8.0_131/jre/lib/security/cacerts

    When selecting a Truststore, ensure that the UI TrustStore settings match or that the selected UI Truststore contains the correct Certificates Authorities (Root CAs or intermidates) used for the Fusion nodes.
    TrustStore Password

    Password for the truststore.

  4. Select to Update once all fields are complete.

    Please note that Client to Fusion Server SSL encryption will be enabled by default. If you wish to disable it, please follow the guidance here immediately after selecting Update (i.e. do not proceed to restart cluster or Fusion services).
    Changes must be applied to all servers

    Changes to SSL settings require the same changes to be made manually in the UI of every other WANdisco Fusion node. Updating will also make changes to the cluster core-site via the management endpoint.

  5. You will need to restart designated cluster services in order to make the SSL configuration active. A restart of the Fusion UI, Fusion Server and IHC Server services will also be required.

Enable SSL (HTTPS) for the Fusion UI

The following procedures are used for setting up SSL encryption for the WANdisco Fusion UI, so that it can be accessed via HTTPS on a web browser.

Before you enable use of HTTPS, ensure that all WANdisco Fusion nodes/zones have been installed and configured, although it is not necessary to have inducted the nodes.

The procedure should be followed for each WANdisco Fusion server in your replicated ecosystem, in turn.

Enable HTTPS on all nodes
If you don’t enable HTTPS on some nodes, some information (e.g. graph data) will not be displayed.
  1. Create a Key Store file using keytool, then save the file to a location on each node where the WANdisco Fusion server can read it.

  2. Log in to WANdisco Fusion UI, click on the Settings tab.

  3. Select to display the UI Settings section.

    wdf2.14 settings uisettings01
    Figure 178. UI Settings
    • Use HTTPS - Once selected, the additional properties will be listed below:

      HTTPS Port

      The TCP port that will be used to access the Fusion UI (default 8443).

      Key Store

      Absolute filesystem path to the keystore.
      Example location
      /opt/wandisco/ssl/keystore.ks

      Key Store Password

      Password for the KeyStore.

      Key Alias

      The Alias of the private key.
      Example
      wandisco

      Key Password

      Password for the Key.

      Trust Store Settings

      Truststore settings can be adjusted without having to enable HTTPS on the Fusion UI.

      You must ensure that the same Truststore used for the Fusion Server / IHC Server SSL configuration is used for the Fusion UI, or that the Truststore selected contains the correct Certificate Authorities (Root CAs or intermediates) used for the Fusion hostnames. This is only required if SSL is configured for Fusion Server / IHC Server.
      Use JVM Trust Store

      Default JVM truststore, usually jssecerts or cacerts.
      Example location
      /usr/java/jdk1.8.0_131/jre/lib/security/cacerts

      Use Custom Trust Store

      A custom trust store. If selected, additional entry fields will appear for Trust Store and Trust Store Password.

      Disable Trust Store

      Disable the Trust Store.

    • Stop WANdisco Fusion UI Server if unable to load the Trust Store
      If there is an issue in reading or accessing the Trust Store, the Fusion UI will not attempt to start.

  4. Select to Update once all fields are complete.

    Changes must be applied to all servers

    Changes to SSL settings require the same changes to be made manually in the UI of every other WANdisco Fusion node.

  5. A restart of the Fusion UI service will be required after enabling SSL on the UI.

Client to Fusion server encryption

When enabling SSL for the Fusion Server / IHC Server, the Fusion Client SSL encryption will be enabled by default. If you wish to disable it, follow the guidance here.

When deploying to an environment, consider whether client to Fusion server encryption is required. This may not be desirable if the traffic is over the LAN, and therefore considered secure. It may not be worth the performance overhead that comes with SSL encryption.

Every host in the cluster that has a Fusion Client installed (i.e. likely all nodes) must have a Truststore for validating certificates in the filesystem path specified in the fusion.client.ssl.truststore property. Without this, any client driven HDFS operations may be impeded due to a lack of trust between the client and the Fusion server.

The configuration for Fusion Client SSL is referenced in the cluster core-site, as well as the application.properties for the Fusion node. Listed below are examples of values for each property:

core-site.xml
fusion.client.ssl.enabled=true
fusion.client.ssl.truststore=/opt/wandisco/ssl/wandisco.ks
fusion.client.ssl.truststore.password=bP0L7SY7f/4GWSdLLZ3e+Qj4EyZB0oyEFZ//UEvB/5HisTD89cG5QhLDWIXFDeZhA0Jrm8PxLaxlYMAc0SzXj8oF/L/T86e1zXpZdvYL6EqUYHyhdQkxD0/EYjoZxnxq8WUc3cDhjyp+2a0jdWRqFvnQq9ovE0H6Q+Mq6gDzRwKGxHa5G8FQuUaOlhYXjzZaA5thUfuEXSSfhZx8eMUUzAM427mkWwGUIRQa5ki5IGuXaS2uqvNzj813QMNyD4Nva50U24GKJK322Peg3wDKKlFiqn01riOi18MidEAxqJyBdlOF5XXsLURN16QcuThez6kU3C3RktZDuRghv6nGrQ==
fusion.client.ssl.truststore.type=JKS
/etc/wandisco/fusion/server/application.properties
fusion.client.ssl.enabled=true

The fusion.client.ssl.password will be shown in encrypted format, to learn how to create an encrypted password, please see the setting a password for SSL encryption section.

Disabling Fusion Client SSL encryption

To disable Fusion Client SSL encryption, ensure the steps below are carried out on the Hadoop Cluster and Fusion node(s).

  1. Disable Fusion Client SSL on the Cluster via the Cluster Manager (e.g. Ambari/Cloudera) by adjusting the property below to false.

    core-site.xml

    fusion.client.ssl.enabled=false

  2. Disable Fusion Client SSL on the Fusion node(s) by adjusting the property below to false.

    /etc/wandisco/fusion/server/application.properties

    fusion.client.ssl.enabled=false

  3. Restart all required services on the Cluster.

  4. Restart the Fusion services on the Fusion node(s) (i.e. Fusion UI, Fusion Server & IHC Server).

Note that if the properties mentioned above are not present in the core-site.xml and application.properties , the default value(s) will be false.

Keystores / Truststores
  • The keystore contains private keys and certificates used by SSL servers to authenticate themselves to SSL clients. By convention, such files are referred to as keystores.

  • When used as a truststore, the file contains certificates of trusted SSL servers, or of Certificate Authorities trusted to identify servers. There are no private keys in the truststore.

Most commonly, cert-based authentication is only done in one direction server→client. When a client also authenticates with a certificate this is called mutual authentication.

While all SSL clients must have access to a truststore, it is not always necessary to create and deploy truststores across a cluster. The standard JDK distribution includes a default truststore which is pre-provisioned with the root certificates of a number of well-known Certificate Authorities. If you do not provide a custom truststore, the Hadoop daemons load this default truststore. Therefore, if you are using certificates issued by a CA in the default truststore, you do not need to provide custom truststores. However, you must consider the following before you decide to use the default truststore:

If you choose to use the default truststore, it is your responsibility to maintain it. You may need to remove the certificates of CAs you do not deem trustworthy, or add or update the certificates of CAs you trust. Use the keytool utility to perform these actions.

Useful links in the Knowledgebase:

Security Considerations

Keystores contain private keys. truststores do not. Therefore, security requirements for keystores are more stringent:

  • Hadoop SSL requires that truststores and the truststore password be stored, in plaintext, in a configuration file that- is readable by all.

  • Keystore and key passwords are stored, in plaintext, in a file that is readable only by members of the appropriate group.

These considerations should guide your decisions about which keys and certificates you will store in the keystores and truststores that you will deploy across your cluster.

Keystores should contain a minimal set of keys and certificates. Ideally you should create a unique keystore for each host, which would contain only the keys and certificates needed by the Hadoop SSL services running on the host. Usually the keystore would contain a single key/certificate entry. However, because truststores do not contain sensitive information you can safely create a single truststore for an entire cluster. On a production cluster, such a truststore would often contain a single CA certificate (or certificate chain), since you would typically choose to have all certificates issued by a single CA.

Do not use the same password for truststores and keystores/keys.
Since truststore passwords are stored in the clear in files readable by all, doing so would compromise the security of the private keys in the keystore.
Setting up SSL manually

What follows is a manual procedure for setting up SSL. In most cases it has been superseded by the above Fusion UI-driven method. If you make changes using the following method, you will need to restart the WANdisco Fusion server in order for the changes to appear in on the Settings tab.

Create the keystores / truststores. Every Fusion Server and IHC server should have a KeyStore with a private key entry / certificate chain for encrypting and signing.

Every Fusion Server must also have a truststore for validating certificates in the path specific in “fusion.client.ssl.truststore”. If enabling client to Fusion server encryption, then every Fusion Client must also have the truststore in the specified path.

The keystores and truststores can be the same file and may be shared amongst the processes.

Setting a password for SSL encryption

Use the provided bash script for generating a password. Run the script at the command line, enter a plaintext password, the script then generates and outputs the encrypted version of the entry:

# cd /opt/wandisco/fusion/tools/bin/
# ./encrypt-password.sh
Please enter the password to be encrypted
> ********
bP0L7SY7f/4GWSdLLZ3e+Qj4EyZB0oyEFZ//UEvB/5HisTD89cG5QhLDWIXFDeZhA0Jrm8PxLaxlYMAc0SzXj8oF/L/T86e1zXpZdvYL6EqUYHyhdQkxD0/EYjoZxnxq8WUc3cDhjyp+2a0jdWRqFvnQq9ovE0H6Q+Mq6gDzRwKGxHa5G8FQuUaOlhYXjzZaA5thUfuEXSSfhZx8eMUUzAM427mkWwGUIRQa5ki5IGuXaS2uqvNzj813QMNyD4Nva50U24GKJK322Peg3wDKKlFiqn01riOi18MidEAxqJyBdlOF5XXsLURN16QcuThez6kU3C3RktZDuRghv6nGrQ==
Server-Server or Server-Client

Configure the keystore for each server:

Key Value Default File

ssl.key.alias

alias of private key/certificate chain in KeyStore.

NA

application.properties

ssl.key.password

encrypted password to key

NA

application.properties

ssl.keystore

path to Keystore

NA

application.properties

ssl.keystore.password

encrypted password to KeyStore.

NA

application.properties

Server-to-Server or Server-to-IHC

Configure the truststore for each server:

Key Value Default File

ssl.truststore

Path to truststore

Default

application.properties

ssl.truststore.password

encrypted password to trust store

Default

application.properties

Fusion client configuration Server-Client only

Configure the truststore for each client:

Key Value Default File

fusion.client.ssl.truststore

Path to trust store

NA

core-site.xml

fusion.client.ssl.truststore.password

Encrypted password for trust store

NA

core-site.xml

fusion.client.ssl.truststore.type

Format of trust store - JKS, PKCS12

JKS

core-site.xml

IHC Server configuration (Server-IHC SSL only)

Configure the keystore for each IHC server:

Key Value Default File

ihc.ssl.key.alias

alias of private key/certificate chain in keystore

NA

.ihc

ihc.ssl.key.password

encrypted password to key

NA

.ihc

ihc.ssl.keystore

path to keystore

NA

.ihc

ihc.ssl.keystore.password

encrypted password to keystore

NA

.ihc

ihc.ssl.keystore.type

JKS, PKCS12

JKS

.ihc

Enable SSL

The following configuration is used to turn on each type of SSL encryption:

Key Value Default File Fusion Server - Fusion Server

ssl.enabled

true

false

application.properties

Fusion Server - Fusion Client

fusion.client.ssl.enabled

true

false

application.properties

Fusion Server - Fusion IHC Server

SSL Debug
Variable Name Example Description

ssl.debug

true

Requires a "true" or "false" value. When set to true debugging mode is enabled.

Changes in any of these values require a restart of the relevant Fusion service(s) (i.e. Fusion Server, IHC Server or UI Server). Any invalid value may prevent these services from loading correctly.

Configure SSL for core Hadoop Services

In order to configure SSL for core Hadoop services (such as HDFS, YARN and MapReduce) and more, please follow the guidance documented by the relevant service provider:

SSL configuration properties for Fusion

Please see the Fusion Configuration Properties section (search for ".ssl.") for a list of related configuration properties.

6.8.2. Kerberos

WANdisco Fusion can run on Kerberized Hadoop clusters, with minimal configuration requirements. The following guide runs through the most deployment scenarios.

Existing cluster
If you are installing Fusion into a cluster that is secured with Kerberos, you will need to enable Kerberos during the Fusion installation process. See Kerberos step.
Setting up a new cluster
If you are setting up your platform from scratch, you may elect to enable Kerberos once Fusion is up and running, in which case you can enable Kerberos through Fusion’s web UI.

Look to the security procedures of your particular form of Hadoop:

Before installing on Cloudera
Ensure that the Cloudera Manager database of Kerberos principals is up-to-date.

Running with unified or per-service principal:

Unified
Some Hadoop platforms are Kerberized under a single hdfs user, this is common in Cloudera deployments. For simplicity, this is what we recommend.

  • Generate a keytab for each of your WANdisco Fusion nodes using the hdfs service, for clarification the steps below present a manual setup:

    ktadd -k fusion.keytab -norandkey hdfs/${hostname}@${krb_realm}

Per-service

  • If your deployment uses separate principals for each HDFS service then you will need to set up a principal for WANdisco Fusion.

  • On the KDC, using kadmin.local, create new principals for WANdisco Fusion user and generate keytab file, e.g.:

    > addprinc -randkey hdfs/${hostname}@${krb_realm}
    > ktadd -k fusion.keytab -norandkey hdfs/${hostname}@${krb_realm}

    Copy the generated keytab to a suitable filesystem location, e.g. /etc/wandisco/security/ on the WANdisco Fusion server that will be accessible to your controlling system user, "hdfs" by default.

We don’t recommend storing the keytab in Hadoop’s own Kerberos /etc/hadoop/conf, given that this is overwritten by the cluster manager.
Setting up handshake tokens

By default, handshake tokens are created in the user’s working directories, e.g. /user/jdoe. It is recommended that you create them elsewhere, using the following procedure:

Open the core-site.xml file and add the following property:

<property>
      <name>fusion.handshakeToken.dir</name>
      <value>/wandisco/handshake_tokens</value>
</property>

Handshake tokens for the cluster will be created at the location set by fusion.handshakeToken.dir e.g., if for DC1 you configure the fusion.handshakeToken.dir to be /wandisco/handshake_tokens, then handshake tokens will be written in /wandisco/handshake_tokens/.fusion/.token_$USERNAME where $USERNAME is the username of the user connecting.

Important requirements:

  • All users of the cluster should have the relevant read and write permissions for this location. Applying the correct permissions to this path is important for the security and performance of the cluster.

  • If the path used contains a special URL character, the characters must be encoded. For example, if the desired path is /my dsm&dir, it needs to be set as /my%20dsm%26dir.

If setting the handshake tokens directory to /wandisco/handshake_tokens, you can create and setup the path for secure access as follows:

# Create the handshake token path and required sub-directory
$ hdfs dfs -mkdir /wandisco/handshake_tokens
$ hdfs dfs -mkdir /wandisco/handshake_tokens/.fusion

# Set ownership to the fusion user for both paths
$ hdfs dfs chown fusionuser:hdfs /wandisco/handshake_tokens
$ hdfs dfs chown fusionuser:hdfs /wandisco/handshake_tokens/.fusion

# Set the handshake token path to allow list and access permissions to group and other
$ hdfs dfs -chmod 755 /wandisco/handshake_tokens

# Set the inner directory to provide access for group and other, but not list or write
# The directory stick bit is also set, ensuring only the path owner or superuser may delete an object
# This permissions set ensures no individual user may delete an object that may result in denial of service to another user
$ hdfs dfs -chmod 1711 /wandisco/handshake_tokens/.fusion

Important: Known issue if using MapR
There are known problems if using MapR with FusionHdfs or FusionHcfs configurations. Some required directories are currently missing. You can work around the problem by creating the following directories, then making sure that Yarn and MapR users are added and that they have access to the directories. E.g.,

sudo -u hdfs hadoop fs -mkdir /user/yarn
sudo -u hdfs hadoop fs -chown yarn /user/yarn
sudo -u hdfs hadoop fs -mkdir /user/mapred
sudo -u hdfs hadoop fs -chown mapred /user/mapred
Kerberos Configuration

This chapter provides an overview of the most common approaches you may taken when preparing Kerberos principals to secure Fusion operations:

Basic requirements

The following requirements need to be considered as part of the work required to enable Kerberos.

  • Hadoop system user dealing with requests from Fusion must be capable of impersonation so that Fusion can proxy requests on that user’s behalf - otherwise, the request will be denied. For more information, read about secure impersonation/DoAs.

  • The same underlying system user must be capable of running HDFS file-system operations on paths for which it is not itself have appropriate permissions. E.g.

    1. Creating the .fusion directories inside the root of replicated paths, specifically in cases where the fusion user does not own the directory.

    2. During replication, the ability of Fusion to delete handshake tokens, as created by a client to provide HDFS access (user, fusion user doesn’t usually have permission to the actioning user’s home directory where the tokens are created by default).

    3. During replication & make consistent operations, the ability to read the contents of data on the source cluster that is under replication or make them consistent with the remote cluster. Also, the ability to write to files that it does not own, and to be able to change ownership of the files after they are written to the destination.

The following four options are provided in order of security in that the later approaches use fewer shared components which reduces exposure or are more readily able to revoke access.

Approach 1: Default Principal (HDFS)

Use of the default available HDFS Kerberos principal as already used by the cluster. In HDP, this typically involves the use of the pre-available hdfs.headless.keytab and hdfs principal.

Headless principals are not bound to a specific host, e.g. "SERVICE@REALM.COM".
Service principals are bound to a specific service and host, e.g. "SERVICE/HOSTNAME@REALM.COM".

This principal already maps to the hdfs user on the underlying file-system, it’s a superuser by default. Depending on the nature of any reason to revoke access, this may involve the complete regeneration of a new hdfs keytab for all services that use it.

Configuration requirements:

The easiest option for configuration, when asked to configure Kerberos in the Fusion installation, you supply the Kerberos keytab as /etc/security/keytabs/hdfs.headless.keytab (location may vary, especially in CDH installs).

This principal already maps internally on the cluster to the superuser (hdfs).

Approach 2: Custom Principal with mapping

Custom Kerberos principal with exported keytab, configured to map to the hdfs user on the underlying system.

Due to the use of mapping of the hdfs user, this means it will be a superuser by default.

Depending on the nature of any reason to revoke access, this could be removing the mapping to the hdfs user or regeneration of only the associated principal keytabs.

Configuration requirements:
  1. Requires manual preparation of the Kerberos principal & keytab, along with configuration to map this user to the existing superuser account.

    Assumptions

    Kerberos realm = EXAMPLE.HADOOP

    Superuser = hdfs

    Principal = fusionuser@EXAMPLE.HADOOP

  2. Create the Kerberos principal in the KDC via the kadmin command on the host of the Fusion server, e.g.

    kadmin -p admin/admin@EXAMPLE.HADOOP -q "addprinc -randkey fusionuser@EXAMPLE.HADOOP"
  3. Export the keytab to appropriate directory.

    mkdir -p /opt/keytabs
    kadmin -p admin/admin@EXAMPLE.HADOOP -q "ktadd -kt /opt/keytabs/fusionuser.keytab fusionuser@EXAMPLE.HADOOP"
    chown -R hdfs:hdfs /opt/keytabs
  4. Add an auth_to_local rule in Hadoop configuration (located in the HDFS service) to map the fusionuser@EXAMPLE.HADOOP principal to the local user hdfs.

    Search for "auth_to_local" under the HDFS config in the cluster, add the following at the top of this box.These rules are read in order and the first match applies. The exact order may vary on configuration choice, please review before committing.

    RULE:[1:$1@$0](fusionuser@EXAMPLE.HADOOP)s/.*/hdfs/

    This rule says if the principal matches fusionuser@EXAMPLE.HADOOP then apply this rule. It will perform a sed (regular expression replacement) by matching the entire principal (.*) and replacing it with the user hdfs.

    Ensure this config is saved and deployed to the cluster.

  5. The Fusion installation will configure a proxy user setting. This setting defines that the user through which Fusion is run as can impersonate another user. If the user is hdfs, the rules are as follows (example provided below).

    hadoop.proxyuser.hdfs.hosts=$FUSION_NODE01,$FUSION_NODE02,etc

    Allow the user hdfs to perform impersonation from any host in the cluster. This can be a comma-separated list of hostnames, and should (at least) contain the hostname of the Fusion server(s).

    hadoop.proxyuser.hdfs.groups=*

    Allow the user hdfs to perform an impersonation of users that are members of any Linux group. This can be a comma-separated list that you may need to refine. Typically this will be set as an asterisk (signifying any group).

  6. Fusion will now use the custom principal and keytab, while granting superuser as the hdfs user, along with impersonation.

Approach 3: Custom Principal with mapping/supergroup

Custom Kerberos principal with exported keytab. Configure the user that this principal maps to into the supplementary group associated with the hdfs superuser group setting (thus making it a superuser).

This is the supplementary Linux user group. The configured group is in the hdfs-site.xml file and the property is named dfs.permissions.superusergroup.

Depending on the nature of any reason to revoke access, this could be removing the mapping to the hdfs user or regeneration of only the associated principal keytabs. Removing the supplementary Linux group will also revoke superuser privileges.

Configuration requirements:

This process will require manual preparation of the Kerberos principal & keytab, along with group membership to make this user a superuser.

  1. You need to know the supergroup, which can be found under the HDFS service setting dfs.permissions.superusergroup. The value is typically "hdfs".

  2. Create the Kerberos principal in the KDC via the kadmin command on the host of the Fusion server.

    kadmin -p admin/admin@EXAMPLE.HADOOP -q "addprinc -randkey fusionuser@EXAMPLE.HADOOP"
  3. Export the keytab to the appropriate directory.

    mkdir -p /opt/keytabs
    kadmin -p admin/admin@EXAMPLE.HADOOP -q "ktadd -kt /opt/keytabs/fusionuser.keytab fusionuser@EXAMPLE.HADOOP"
    chown -R fusionservice:hdfs /opt/keytabs
  4. Add an auth_to_local rule in Hadoop configuration (located in the HDFS service) to map the fusionuser@EXAMPLE.HADOOP principal to the local user hdfs.

    Search for auth_to_local under the HDFS config in the cluster. Add the following at the top of this box, this will map it to the current non-superuser fusionservice.

    RULE:[1:$1@$0](fusionuser@EXAMPLE.HADOOP)s/.*/fusionservice/

    These rules are read in order and the first match applies. The exact order may vary on configuration choice, please review before committing.

    This rule says if the principal matches fusionuser@EXAMPLE.HADOOP then apply this rule. It will perform a sed (regular expression replacement) by matching the entire principal (.*) and replacing it with the user fusionservice.

    Ensure this config is saved and deployed to the cluster.

  5. You must create Linux user fusionservice on all hosts in the cluster, i.e.

    adduser fusionservice
  6. To make this user a superuser in HDFS, ensure this user is a member of the superusergroup. This can be a supplementary group and does not need to be the primary group.

    usermod -a -G hdfs fusionservice
  7. The Fusion installation will configure a proxyuser setting. This setting defines that the user through which Fusion is run as can impersonate another user. In this case, fusionservice (example provided below):

    hadoop.proxyuser.fusionservice.hosts=$FUSION_NODE01,$FUSION_NODE02,etc

    Allow the user fusionservice to perform impersonation from any host in the cluster. This can be a comma-separated list of hostnames, and should (at least) contain the hostname of the Fusion server(s).

    hadoop.proxyuser.fusionservice.groups=*

    Allow the user fusionservice to perform an impersonation of users that are members of any Linux group. This can be a comma-separated list that you may need to refine. Typically this will be set as an asterisk (signifying any group).

  8. Fusion will now use the custom principal and keytab, while granting superuser as the fusionservice user through the supergroup, along with impersonation.

This configuration can be made even more secure by creating a user that is in the supergroup, and then pointing fusion.system.user at that user instead of hdfs itself. This additional level of security / auditing capability brings this approach in line with Approach 4.
Approach 4: Custom Principal with mapping/supergroup and additional auditing capability

Create a custom Kerberos principal with exported keytab, then use the core-site.xml parameter fusion.system.user to point an actual HDFS superuser. Fusion will then proxy to that user when required. This method provides auditing benefits as all superuser actions by Fusion are visible in the hdfs-audit.log, which will report each proxy request and who proxied who.

Configuration requirements:

This process will require manual preparation of the Kerberos principal & keytab, along with group membership to make this user a superuser.

  1. You need to know the supergroup, which can be found in the HDFS service setting dfs.permissions.superusergroup. The value is typically "hdfs" but may vary.

  2. Create the Kerberos principal in the KDC via the kadmin command on the host of the Fusion server.

    kadmin -p admin/admin@EXAMPLE.HADOOP -q "addprinc -randkey fusionuser@EXAMPLE.HADOOP"
  3. Export the keytab to the appropriate directory.

    mkdir -p /opt/keytabs
    kadmin -p admin/admin@EXAMPLE.HADOOP -q "ktadd -kt /opt/keytabs/fusionuser.keytab fusionuser@EXAMPLE.HADOOP"
    chown -R fusionservice:hdfs /opt/keytabs
  4. Add an auth_to_local rule in Hadoop configuration (located in the HDFS service) to make the fusionuser@EXAMPLE.HADOOP to be the local user "hdfs".

    Search for auth_to_local under the HDFS config in the cluster. Add the following at the top of this box, this will map it to the current non-superuser fusionservice.

    RULE:[1:$1@$0](fusionuser@EXAMPLE.HADOOP)s/.*/fusionservice/

    These rules are read in order and the first match applies. The exact order may vary on configuration choice, please review before committing.

    This rule says if the principal matches fusionuser@EXAMPLE.HADOOP then apply this rule. It will perform a sed (regular expression replacement) by matching the entire principal (.*) and replacing it with the user fusionservice.

    Ensure this config is saved and deployed to the cluster.

  5. You must create Linux user fusionservice on all hosts in the cluster, i.e.

    adduser fusionservice
  6. The Fusion installation will configure a proxyuser setting. This setting defines that the user through which Fusion is run as can impersonate another user. In this case, fusionservice (example provided below):

    hadoop.proxyuser.fusionservice.hosts=$FUSION_NODE01,$FUSION_NODE02,etc

    Allow the user fusionservice to perform impersonation from any host in the cluster. This can be a comma-separated list of hostnames, and should (at least) contain the hostname of the Fusion server(s).

    hadoop.proxyuser.fusionservice.groups=*

    Allow the user fusionservice to perform an impersonation of users that are members of any Linux group. This can be a comma-separated list that you may need to refine. Typically this will be set as an asterisk (signifying any group).

  7. Edit the core-site.xml (via the appropriate manager UI) and add the following property:

    fusion.system.user=hdfs

    where "hdfs" is a superuser.

  8. Deploy the configs and restart the Fusion processes. Fusion will now use the custom principal and keytab, while granting superuser privileges by way of proxying to the fusion.system.user.

Kerberos and HDP’s Transparent Data Encryption

There are some extra steps required to overcome a class loading error that occurs when WANdisco Fusion is used with at-rest encrypted directories. Specifically for Ranger Key Management Service (Ranger KMS), cluster configuration changes as per the example below:

<property>
<name>hadoop.kms.proxyuser.fusionuser.users</name>
<value>*</value>
</property>

<property>
<name>hadoop.kms.proxyuser.fusionuser.groups</name>
<value>*</value>
</property>

<property>
<name>hadoop.kms.proxyuser.fusionuser.hosts</name>
<value>*</value>
</property>

The fusionuser is the example provided for whichever user Fusion is mapped to (e.g. from auth_to_local translation).

API with Kerberos enabled

If you have Kerberos-authentication enabled on REST API, you must kinit before making REST calls, and enable GSS-Negotiate authentication. To do this with curl, you must include the --negotiate and -u: options e.g.

curl --negotiate -u: -X GET "http://${HOSTNAME}:8082/fusion/fs/transfers"
Kerberos Troubleshooting

This section covers some recommended fixes for potential Kerberos problems.

Kerberos Error with MIT Kerberos 1.8.1 and JDK6 prior to update 27

Prior to JDK6 Update 27, Java fails to load the Kerberos ticket cache correctly when using MIT Kerberos 1.8.1 or later, even after a kinit.

The following exception will occur when attempting to access the Hadoop cluster.

WARN ipc.Client: Exception encountered while connecting to the server : javax.security.sasl.SaslException:
GSS initiate failed [Caused by GSSException: No valid credentials provided (Mechanism level: Failed to find any Kerberos tgt)]

The workaround is:

Error "Can’t get Kerberos realm" when installing WANdisco Fusion.

WANdisco Fusion uses the settings that are written in the krb5.conf file to configure Kerberos. The default realm (default_realm) is one of the values that must be specified. If not, then the JVM will fallback by trying to get the default realm through DNS. If this fails, then you see the "Can’t get Kerberos realm" error message.

Workaround

The workaround is to properly configure the default_realm in krb5.conf file.

6.8.3. LDAP/ Active Directory

Set up LDAP

See the LDAP Settings section for guidance on setting up LDAP/AD authentication for the Fusion UI.

6.8.4. Roles and Permissions

When LDAP is enabled the Fusion UI will list a new section Roles and Permissions under the Security section of the Settings tab.

To assign roles and control the permissions available to each role, click on the Roles and Permissions button.

LDAP roles
Figure 179. Update LDAP Settings

By default, Fusion comes with the following defined roles.

Role Priority See Role Priority

Audit

4

Content

3

Infrastructure

2

Superuser

1

Audit

The Audit role is intended for access that requires no actual interaction. This role lets an account get access to the Fusion UI without actually being able to make any changes. By default, the Audit user has the lowest priority (4).

Audit role permission
  • Has READ access to everything.

  • No write access of any sort.

Content

The Content role provides access to the basic Fusion UI functionality, allowing a user with this role to modify replication rules, without having access to deeper configuration-related functions.

Content role permissions
  • Has WRITE access to Replication Rules (eg CRUD)

  • Has WRITE access to Consistency Checks and Make Consistent operations (eg can trigger these)

  • Has READ access to the Dashboard

  • Has NONE access to the Nodes page

  • Has READ access to the Settings page

  • Has NONE access to any feature within this other than Email Notifications

Infrastructure

The Infrastructure role provides access to a system administrator level user.

Infrastructure role permissions
  • Has WRITE access to all system or performance related features (eg Dashboard, Nodes page, Settings page)

  • Has READ access to everything else (for diagnostic purposes).

  • Writer node information, which is required for the infrastructure role, will be added onto the node profile page

Superuser

The Super User has all complete access. For this reason, it is not intended to be used for regular operation. The Superuser has the highest priority (1).

Superuser role permissions
  • Has WRITE access to everything, permissions can not be modified.

You can reset the Superuser password using the following - Generate a new password
Role Priority

Each role is assigned a priority parameter (int value). The priority for a role must be unique, means there should be no roles with the same priority.

In the event that a user account has been assigned multiple roles, the priority is used to set the order in which Fusion checks for applicable permissions. This check is done each time the applicable user logs in, in which case, changes to a user’s role and permissions will not be picked up until they next log in. For example, a user assigned with the Super User role will always be checked first, and it’s permissions will apply over any other roles with more limited permissions.

Role Permissions limit access to UI Features
This user guide is written to provide the most complete view of the Fusion UI, as if you are logging in with the Superuser or equivalent role. If you access the Fusion UI using an account with an assigned role that has a limited set of permissions, then you many not see some features or may not permit you to edit those features.

The Fusion UI will limit access to those functions and features that match the role and permissions of the logged-in account. For example, logging in with a Super User role will provide unlimited access, while a user logging in with an account assigned with the Audit role will only have access to a very limited feature-set.

The permissions available to each role can be customised in this section. Consider creating new roles if you need a bespoke set of permissions, so that you don’t lose track of unexpected access or restrictions that have been applied to the default roles.
Add new role

The Add new role function lets you create a brand new role with a custom set of permissions. Use this feature if you need to assign users to roles that do not match with those roles that are provided by default.

LDAP Roles screen
Figure 180. Create a new role
Fusion role

Role properties for the new role.

Name

a unique name for the role

Priority

a unique priority number. see Role Priority for more guidance on how priority is used.

LDAP users for

LDAP search properties for the new role.

Search Base

the starting point in your LDAP for user searches

Search Filter

the filter used for user searches

Permissions for

Permissions that will be assigned to the new role. Tick a checkbox to provide Read and or Write access. If neither Read or Write are ticked then this role will get no access to that resource.

Replication

Permissions for access to Consistency check, Create rules, Edit rules, Remove rules, Make Consistent

Plugins

Permissions for enabled plugins

Global settings

ADLS settings, AWS credentials, Azure settings, Bandwidth limits, Chunk size, Client bypass, Consistency check settings Core http settings, DConE database, EMR client, EMR settings, Fusion roles, Google settings, Graph settings, Heap size, IHC server, Kerberos, LDAP settings, License, Log settings, Networking, Notifications, Plugin installation, Push threshold, Replication settings, S3 settings, S3 throttle, Swift settings, UI server http settings, URI selection

Infrastructure

Induction, Node operations, User management, View logs

Resources

Client download, Fusion assets download

System information

About versions, Aggregate throughput, Fusion db disk usage, System CPU load, System status

Support

Contact information, License information

Edit role
LDAP Roles screen
Figure 181. Create a new role

Select from the existing roles, then click the Edit role. You can make modifications to any of the role’s properties. Click Update to save any changes you have made or click Reset to reload the exiting permissions.

View role

Select from the existing roles, then click the View role button. This is a read-only version of the Edit role screen. You can’t make changes here, instead used the Edit role screen.

Delete role

Select from the existing roles, then click the Delete role button. This will

  1. Click on the Delete role button.

    LDAP Roles screen
    Figure 182. Delete a role
  2. The following message will appear:

    Are you sure you want to delete the selected role?

    Click Confirm to continue with the deletion.

    LDAP Roles screen
    Figure 183. Delete a role
  3. The deletion will be confirmed with the message:

    Role deleted successfully

    Click Close.

    LDAP Roles screen
    Figure 184. Confirm delete of a role

6.9. Troubleshooting

This section details with how to diagnose and fix problems that many occur in deployment. It’s important that you check the Release Notes for any Known issues in the release that you are using.

6.9.1. Read the logs

There are a number of log files that provide information that will be necessary in finding the cause of many problems.

The log files for WANdisco Fusion are spread over three locations. Some processes contain more than one log file for the service. All pertinent log files are captured by running the WANdisco talkback shell script that is covered in the next section.

Log Directory Settings
Log Settings

On the Settings tab of the Fusion UI, you will find a Log Directory Settings screen. The screen defines the log directories of the Fusion Core Server, Fusion IHC Server and Fusion UI. The Fusion user must have full permissions on these directories. See Configure log directory via UI.

WANdisco Fusion Server Logs

The logs on the WANdisco Fusion server record events that relate to the data replication system.

Log locations

/var/log/fusion/server

Primary log(s)

fusion-server.log
This is the live log file for the running Fusion server process.

fusion-server.log.yyyy-mm-ddTHH:MM:SS
Rotation is presently defaulted at 200MB with a retention of 100 files, although this can be customised. The log will also rotate after a restart.

Historical logs

The following logs are listed for completeness but are not generally useful for monitoring purposes.

http_access.log
http_access.log.yyyy-mm-ddTHH:MM:SS
This is a log of all requests recevied by the Fusion server API endpoint.

gc.log.yyyy-mm-dd_HHMMSS.0.current
Garbage Collection logging of the Fusion server Java process.

WANdisco Fusion UI Server Logs

The WANdisco Fusion user interface layer, responsible for handling interactions between the administrator, WANdisco Fusion and the Hadoop Management layer.

Log locations

/var/log/fusion/ui/

Primary log(s)

fusion-ui.log

Historical logs

fusion-ui.log.x

The UI logs will contain errors such as failed access to the user interface, connectivity errors between the user interface and WANdisco Fusion Server REST API and other syntax errors between the user interface and the WANdisco Fusion server’s REST API and other syntax errors whilst performing administrative actions across the UI.

Inter-Hadoop Connect (IHC) Server Logs

Responsible for streaming files from the location of the client write to the WANdisco Fusion server process in any remote cluster to which hadoop data is replicated.

Log location

/var/log/fusion/ihc
/var/log/fusion/ihc/server

Primary log(s)

server/fusion-ihc-ZZZ-X.X.X.log

  • The live IHC process log files. The components of the filename are as follows:
    ZZZ - Hadoop distribution marker (HDP, CDH, etc). This will be "hdp" for a Hortonworks integrated cluster.
    X.X.X - A matching cluster version number. This will be "2.6.0" for a Hortonworks 2.6 cluster.

Historical logs

server/fusion-ihc-ZZZ-X.X.X.log.yyy-mm-dd
log_out.log
This log file contains details of any errors by the process when reading from HDFS in the local cluster, such as access control violations, or network write errors when streaming to the WANdisco Fusion server in any remote cluster.

WANdisco Fusion Client Logging

By default, the WANdisco Fusion client remains silent and will not provide an indication that it has been loaded or is in use by an application. For troubleshooting purposes, it can help to enable client logging to allow you to determine when the client is in effect. Client logging can be enabled by adding an entry to the cluster’s log4j.properties file similar to that below:

log4j.logger.com.wandisco.fs.client=INFO

Once enabled, client log information will be produced by default on the standard output. Either removing this entry, or setting the logging level for the WANdisco Fusion client library to "OFF" will restore default behavior with no client-side logging.

Log analysis

This is the standard format of the WANdisco log messages within Fusion. It includes an ISO8601 formatted timestamp of the entry, the log level / priority, followed by the log entry itself. Log levels we provide in order of severity (highest to lowest) that you may observe:

  • PANIC

  • SEVERE

  • ERROR

  • WARNING

  • INFO

For log analysis and reporting, logs with at the PANIC, SEVERE and ERROR levels should be investigated. The warning level messages indicate an unexpected result has been observed but one that hasn’t impacted the system’s continued operation. Additional levels may exist, but are used in cases when the logging level has been increased for specific debug purposes. At other times, other levels should be treated as informational (INFO).

Quickly picking out problems

One simple thing that can be done is to grep the log file for any instance of "exception" and/or "PANIC" - this will tell the administrator a great deal without much effort. Using something like:

cat /var/log/fusion/server/fusion-dcone.log.0 | egrep -i "exception|panic"

6.9.2. Failed To Publish Your Settings

During Initial Configuration, settings need to be published to take effect.

If you get the warning Failed To Publish Your Settings, the most likely reason is that the Cluster Manager configuration has not been updated. To fix this, repeat the Cluster Manager configuration steps:

  1. On the Settings page go to Hadoop → Cluster Manager and fill in the relevant fields.

  2. Click Validate. If validation fails, ensure any corrections are made.

  3. Click Update.

  4. Re-attempt to Publish.

If these steps are not successful, please contact WANdisco support.

6.9.3. About this Node

Under the Settings tab there is an About screen that provides the following information, useful for quickly comparing the versions of different nodes in a deployment.

Fusion13
Figure 185. About

The About This Node panel shows the version information for the underlying Hadoop deployment as well as the WANdisco Fusion server and UI components:

Fusion UI Version

The current version of the WANdisco Fusion UI.

Fusion Build Number

The specific build for this version of the WANdisco Fusion UI.

Hadoop Version

The version of the underlying Hadoop deployment.

Fusion Version

The version of the WANdisco Fusion replicator component.

Fusion Uptime

The time elapsed system the WANdisco Fusion system last booted up.

Cluster Manager

The management application used with the underlying Hadoop.

6.9.4. Support

The support tab contains links and details that may help you if you run into problems using WANdisco Fusion.

WANdisco Fusion tree
Figure 186. Support tab

6.9.5. Talkback

Talkback is a bash script that is provided in your WANdisco Fusion installation for gathering all the logs and replication system configuration that may be needed for troubleshooting problems. Should you need assistance from WANdisco’s support team, they will ask for an output from Talkback to begin their investigation.

Talkback location

You can find the talkback script located on the WANdisco Fusion server’s installation directory:

cd /opt/wandisco/fusion/tools/bin/

You can run talkback as follows:

./talkback.sh

If a cluster has Kerberos security enabled (Talkback will detect this from WANdisco Fusion’s configuration), you may be asked for Kerberos details needed to authenticate with the cluster.

For more information on talkback usage you can run:

./talkback.sh -h

Talkback can be run interactively or non-interactively.

To run non-interactively the following variables must be set. If you do not set these variables you will be prompted.

  • FUSION_KERBEROS_CONFIGS - Set to "true" or "false" to grab the Kerberos configs (not necessary if Kerberos is not enabled).

  • FUSION_PERFORM_FSCK - Set to "true" or "false" to perform a file system consistency check.

    • If running interactively you will be asked if you wish to perform a HDFS fsck, or not. Option 1 for yes, option 2 for no.

  • FUSION_TALKBACK_DIRECTORY - Set the absolute path directory where the tarball will be saved.
    Note, WANdisco Fusion talkbacks can exceed 300MB compressed, but well over 10GB uncompressed (due to logs).

The following variables can also be set:

  • FUSION_JSTACK - Set to "true" or "false" to run JStack. Defaults to "false".

  • FUSION_KERBEROS_ENABLED - Set to "true" or "false" on whether Kerberos enabled. Talkback will check if left null.

  • FUSION_LOG_PERIOD - Set to positive integer on how long to set timeout for log copies. Defaults to 15.

  • FUSION_MARKER - Set to include custom marker inside of talkback filename. Defaults to "FUSION" if left null.

  • FUSION_PMAP - Set to "true" or "false" to grab the pmap of the file. Defaults to "false".

  • FUSION_PROC - Set to "true" or "false" to grab the proc of the file. Defaults to "true".

  • FUSION_PROXY_USER - Set the proxy user for curls. Defaults to null.

  • FUSION_SUPPORT_TICKET- Set ticket number to give to WANdisco support team. Defaults to null.

  • FUSION_TIMEOUT - Set the timeout of the CURL commands. Defaults to two minutes.

  • DFS_COUNT - Set to "true" or "false" to run 'hdfs dfs -count /repl' on HDFS distros. Defaults to false.

  • IS_HADOOP_ZONE - Set to "true" or "false" depending on whether Hadoop distro (only necessary if cannot detect distro).

  • HADOOP_RETRY - Set to "true" or "false" to bypass Hadoop host prompts.

  • SSH_USER_KERBEROS - SSH to KDC as a different user. Defaults to null.

  • TALKBACKNAME - Set the talkback filename to something other than the default format.

PID Variables: Used for JStack, PMap and proc file capture:

  • FUSION_SERVER - Set to "true" or "false" to capture pid info for fusion server. Defaults to "true".

  • FUSION_IHC_SERVER - Set to "true" or "false" to capture pid info for ihc server. Defaults to "true".

  • FUSION_UI_SERVER - Set to "true" or "false" to capture pid info for UI server. Defaults to "false".

Running talkback

To run the talkback script, follow this procedure:

  1. Log in to the Fusion server. If you’re not logged in as root, use sudo to run the talkback script. Below is an example output:

          ====================== INFO =========================
          The talkback agent will capture relevant configuration
          and log files to help WANdisco diagnose the problem
          you may be encountering. Use '-h' flag for more info.
    
    Retrieving current system state information
    Capturing uptime.
    Capturing pwd.
    Capturing uname -a.
    Capturing df -k.
    Capturing df -k /opt/wandisco/fusion/server/dcone.
    Capturing mount.
    Capturing /usr/java/jdk1.7.0_79/bin/java -version.
    Capturing /bin/bash -version.
    /usr/bin/lsb_release
    Capturing rpm -qa | grep coreutil.
    Capturing ps -leaf | grep java.
    Capturing ps -C java -L -o tid,pcpu,time.
    Capturing ps -leaf.
    Capturing top -b -n 1.
    Capturing netstat -anp.
    Found 'fusion-server' with user: 'hdfs' and pid: '26950'.
    Capturing proc file for 'fusion-server'.
    Found 'fusion-ihc-server' with user: 'hdfs' and pid: '24536'.
    Capturing proc file for 'fusion-ihc-server'.
    Copying Fusion UI log files, this can take several minutes.
    Gathering information from Fusion endpoints
    Protocol is:  http
    Hostname is:  <your.hostname>
    Port is:  8082
    retrieving details for node <your.nodeID>
    Capturing ping -c 5 <your.hostname>.
    Copying Fusion server log files, this can take several minutes.
    Copying Fusion IHC log files, this can take several minutes.
    WARNING: fs.xml could not be parsed. This is expected behavior when no replicated directories exist.
    Gathering Java Management Extension data.
    
    2 archives were successfully processed.
    
    2 archives were successfully processed.
    Would you like to include hadoop fsck? This can take some time to complete and may drastically increase the size of the tarball.
    1) Yes
    2) No
    ? 1
    Capturing sudo -u hdfs hadoop --config /etc/hadoop/conf fsck / -blocks -locations -racks -files -openforwrite.
    Running sysinfo script to capture maximum hardware and software information...
    Gathering Summary info....
    Gathering Kernel info....
    Gathering Hardware info....
    Gathering File-Systems info....
    Gathering Network info....
    Gathering Services info....
    Gathering Software info....
    Gathering Stats info....
    Gathering Misc-Files info....
    THE FILE sysinfo/sysinfo_<your.hostname>-20171108-100136.tar.gz HAS BEEN CREATED BY sysinfo
    
    TALKBACK COMPLETE
    
    ---------------------------------------------------------------
     Please upload the file:
    
         /<your specified directory>/talkback-FUSION-201711081510-<your.hostname>.tar.gz
    
     to WANdisco support with a description of the issue.
    
     Note: do not email the talkback files, only upload them
     via ftp or attach them via the web ticket user interface.
    --------------------------------------------------------------
  2. Follow the instructions for uploading the output on WANdisco’s support website.

Uploading talkback files

If you need help from WANdisco support you may need to send them your talkback output files.
DO NOT send these files by email. The best way to share your talkback files is via SFTP, but small files (<50MB) can also be uploaded directly at customer.wandisco.com.

For information on how to upload talkback files, see the Knowledge base article Uploading Logs to the Customer FTP Space.

Information can also be found at customer.wandisco.com but you will need a valid WANdisco License Key to access this information.

6.9.6. Common problems

Moving objects between mismatched filesystems

If you move objects onto the distributed file system you must make sure that you use the same URI on both the originating and destination paths. Otherwise you’d see an error like this:

[admin@vmhost01-vm1 ~]$ hadoop fs -mv /repl2/rankoutput1 fusion:///repl2/rankoutput2/
15/05/13 21:22:40 INFO client.FusionFs: Initialized FusionFs with URI: fusion:///, and Fs: hdfs://vmhost01-vm1.cluster.domain.com:8020. FileSystem: DFS[DFSClient[clientName=DFSClient_NONMAPREDUCE_-721726966_1, ugi=admin@DOMAIN.EXAMPLE (auth:KERBEROS)]]
mv: `/repl2/rankoutput1': Does not match target filesystem

If you use the fusion:/// URI on both paths it will work, e.g.

[admin@vmhost01-vm1 ~]$ hadoop fs -mv fusion:///repl2/rankoutput1 fusion:///repl2/rankoutput1
15/05/13 21:23:27 INFO client.FusionFs: Initialized FusionFs with URI: fusion:///, and Fs: hdfs://vmhost01-vm1.cluster.domain.com:8020. FileSystem: DFS[DFSClient[clientName=DFSClient_NONMAPREDUCE_-1848371313_1, ugi=admin@DOMAIN.EXAMPLE (auth:KERBEROS)]]

Note that since the non-replicated directory doesn’t yet exist in ZONE2 it will get created without the files it contains on the originating zone. When running WANdisco Fusion using the fusion:///, moving non-replicated directory to replicated directory will not work unless you use of the fusion:/// URI.

You can’t move files between replicated directories
Currently you can’t perform a straight move operation between two separate replicated directories.

Handling file inconsistencies

WANdisco Fusion’s replication technology ensures that changes to data are efficiently propagated to each zone. However, the replication system is optimized for maintaining consistency through transactional replication and is not designed to handle the initial synchronization of large blocks of data. For this requirement, we have the Consistency Check tool.

Transfer reporting

When looking at the transfer reporting, note that there are situations in which HFlush/early file transfer where transfer logs will appear incorrect. For example, the push threshold may appear to be ignored. This could happen if an originating file is closed and renamed before pulls are triggered by the HFlush lookup. Note that although this results in confusing logs, those logs are in fact correct; you would see only two appends, rather than the number determined by your push threshold - one in the very beginning, and one from the rename, which pulls the remainder of the file. What is happening is optimal; all the data is available to be pulled at that instant, so we might as well pull all of it at once instead of in chunks.

7. Reference Guide

7.1. Fusion Configuration Properties

This section lists the available configuration for WANdisco Fusion’s component applications.

You should take care when making any configuration changes on your clusters, always check with WANdisco support before changing properties in your environment.

This section describes the configuration files and the properties editable in them.

  • /etc/wandisco/fusion/server/application.properties - contains WANdisco Fusion Server properties

  • /etc/wandisco/fusion/ihc/server/{distro}/{version string}.ihc - contains all the IHC server properties

  • config.properties - contains all the Fusion Client properties (see note below)

  • core-site.xml - contains WANdisco Fusion Client and WANdisco Fusion Server properties

From 2.14, an additional file can be used to configure the Fusion Client - config.properties. Using this file you can make updates to Fusion Client properties without editing the core-site.xml, which would require a restart of Hadoop services.

7.1.1. Fusion Server configuration

The Fusion Server configuration properties are found in:

/etc/wandisco/fusion/server/application.properties

All these properties are used by the WANdisco Fusion Server.

Table 4. Fusion Server properties
Property Description

agreementstore.segment.size

The DConE AgreementStore stores agreements in segments, with each segment stored in its own file. This defines how many agreements are stored in a segment.

Permitted value Default Checked at

1 - MAX_INTEGER

5000

Startup

announcement.key.interval

Specifies the minimum Global Sequence Number (GSN) interval that triggers the generation of a status message.

Permitted value Default Checked at

1 - MAX_LONG

1000L

Dynamic

announcement.time.interval

Specifies the minimum time interval that triggers the generation of a status message.

Permitted value Default Checked at

1 - MAX_LONG

10000L

Dynamic

application.hostname

This is the hostname used in reporting the Fusion server’s address.

Permitted value Default Checked at

A valid hostname

None

Startup

application.integration.db.deep.copy

If set to true, the application integration database will also write to disk rather than being exclusively stored in memory.

Permitted value Default Checked at

True/false

False

Startup

application.integration.db.force

If set to true, the application integration database will forcibly sync with the filesystem.

Permitted value Default Checked at

True/false

False

Startup

application.integration.db.panic.if.dirty

If set to true and the application integration database was not shut down 'cleanly', then on restart the server will not start.

Permitted value Default Checked at

True/false

False

Startup

application.location

The directory the application will use for persistence.

Permitted value Default Checked at

Any existing path

.

Startup

application.port

The port DConE uses for communication.

Permitted value Default Checked at

1 – 65535

6444

Startup

communication.hostname

This is the hostname used for binding opened ports (for DConE, requests port, REST).

Permitted value Default Checked at

A valid hostname

0.0.0.0

Startup

connection.reset.count

The number of message objects stored in cache before the cache is reset.

Permitted value Default Checked at

1 - MAX_INTEGER

5000

Startup

database.location

The directory DConE will use for persistence.

Permitted value Default Checked at

Any existing path

/opt/wandisco/fusion/server/dcone/db

Startup

dcone.system.db.deep.copy

If set to true, the DConE database will also write to disk rather than being exclusively stored in memory.

Permitted value Default Checked at

True/false

False

Startup

dcone.system.db.force

If set to true, the DConE database will forcibly sync with the filesystem.

Permitted value Default Checked at

True/false

False

Startup

dcone.system.db.panic.if.dirty

If set to true/false and the DConE system database was not shut down 'cleanly', then on restart the server will not start.

Permitted value Default Checked at

True/false

False

Startup

dcone.teach.limit

Specifies the maximum number of agreements sent in a teach message.

Permitted value Default Checked at

1 - MAX_INTEGER

5000

Dynamic

dcone.use.boxcar

Whether the DConE database should use boxcars or not.

Permitted value Default Checked at

True/false

False

Startup

decoupler.pool.size

Specifies the size of the default decoupler’s thread pool.

Permitted value Default Checked at

1 - MAX_INTEGER

100

Startup

decoupler.queue.size.max

Maximum number of queued items for the decoupler.

Permitted value Default Checked at

1 - MAX_INTEGER

1000000

Startup

decoupler.teach.pool.size

Specifies the size of the teach decoupler’s thread pool; if 0, disables the teach decoupler.

Permitted value Default Checked at

1 - MAX_INTEGER

20

Startup

decoupler.teach.queue.size.max

Maximum number of queued items for the teach decoupler.

Permitted value Default Checked at

1 - MAX_INTEGER

1000000

Startup

executor.threads

The number of threads executing agreements in parallel (this is total number of repair and agreement execution threads).
The names of the threads start with FastExecutor or AgreedExecutor. These can be used to determine where a request was executed and what type of request - the Completed log message of request contains the thread name, which would either start with AgreedExecutor or with FastExecutor.
The fast requests are all requests that do not involve a pull of data from another zone. An agreed request is the opposite of this, whereby data is pulled from another zone as part of the request.

Permitted value Default Checked at

1 – reasonable max number of threads as allowed per platform (taking into account other threads)

250

Startup

fusion.client.ssl.enabled

Enables SSL between the Fusion Server - Fusion Client.

Permitted value Default Checked at

Usage

True/false

False

fusion.http.authentication.enabled

Enables authentication on the REST API.

Permitted value Default Checked at

True/false

False

Startup

fusion.http.authentication.kerberos.keytab

The path to a keytab that contains the principal specified in fusion.http.authentication.kerberos.principal.

Permitted value Default Checked at

Absolute path to Kerberos keytab

N/A

Startup

fusion.http.authentication.kerberos.principal

The principal the fusion server will use to login with. The name of the principal must be "HTTP".

Permitted value Default Checked at

Kerberos principal in HTTP/_HOST@{KERBEROS_REALM} format

N/A

Startup

fusion.http.authentication.simple.anonymous.allowed

If type is "simple", whether anonymous API calls are allowed. If set to false, users must append a query parameter at the end of their URL user.name=$USER_NAME.

Value Checked at Usage

True/false

Startup

Server

fusion.http.authentication.type

Type of authentication to use for http access.

Permitted value Default Checked at

simple or kerberos

simple

Startup

fusion.http.authorization.authorized.proxies

Users that are allowed to proxy on behalf of other users. HTTP calls would include a value for the header proxy.user.name. The proxied user’s permissions will then be checked against authorized readers and read-writers.

Permitted value Default Checked at

Comma-delimited list of users

HTTP

Startup

fusion.http.authorization.authorized.readers

Users that are allowed to make read calls ONLY (write calls are PATCH, POST, PUT, DELETE).

Permitted value Default Checked at

Comma-delimited list of users

fusionUISystem

Startup

fusion.http.authorization.authorized.read.writers

Users that are allowed to make read OR write calls (any type of HTTP request).

Permitted value Default Checked at

Comma-delimited list of users

fusionUISystem

Startup

fusion.http.authorization.enabled

Enables authorization on the REST API. Authentication must also be enabled.

Permitted value Default Checked at

True/false

False

Startup

fusion.http.policy

Determines the transfer protocol(s) to be supported by Fusion Server.

Permitted value Default Checked at

HTTP_ONLY, HTTPS_ONLY, BOTH_HTTP_HTTPS

HTTP_ONLY

Startup

fusion.replicated.dir.exchange

Location of a directory in the replicated filesystem to which Fusion server will write information about replicated directories for clients to read. It’s necessary to configure the same in the core-site.xml, so that it generates the necessary data.

Permitted value Default Checked at

Underlying filesystem path not under replication

/wandisco/exchange_dir

Startup

ihc.ssl.enabled

Whether SSL is enabled for IHC network communications.

Permitted value Default Checked at

True/false

False

Startup

jetty.http.port

The port the Fusion HTTP API will use.

Permitted value Default Checked at

1 – 65535

8082

Startup

jetty.https.port

The port the Fusion HTTPS API will use (if SSL is enabled).

Permitted value Default Checked at

1 – 65535

8084

Startup

learner.max.output.lag

The maximum agreements that a node in a Zone can be behind before it is disabled (i.e. it will no longer be able to propose changes).

Permitted value Default Checked at

1 - MAX_LONG

10000000L

Startup

log.location

The directory in which the Fusion Server stores log files.

Permitted value Default Checked at

A local filesystem directory

/var/log/fusion/server

Startup

maximum.size

The number of agreements DConE will hold in the agreement store - this store holds agreements to be processed and agreements stored to teach other nodes.

Permitted value Default Checked at

1 - MAX_INTEGER

50000

Startup

node.id

The unique identifier given to the Fusion node automatically at installation.

Permitted value Default Checked at

A Hexadecimal number (must not be altered)

N/A

Startup

node.name

The name set for the Fusion node. Adjustable in the Fusion UI.

Permitted value Default Checked at

Any String

N/A

Startup

remote.ihc.port

The port remote ihc servers should connect to when the zone is set to Inbound connection.

Permitted value Default Checked at

1 – 65535

8024

Startup

repair.thread.limit

Maximum number of outstanding files that a single repair will have scheduled for execution at any given time. This is a mechanism for allowing multiple parallel repairs to run together.

Permitted value Default Checked at

1 - MAX_INTEGER

25

Startup

repair.threads

Number of executor threads dedicated for repair tasks only.

Permitted value Default Checked at

1 - MAX_INTEGER (must be less then executor.threads value)

50

Startup

request.port

The port Fusion clients will use to connect the Fusion server.

Permitted value Default Checked at

1 – 65535

8023

Startup

ssl.enabled

Whether SSL is enabled for Fusion Server communications.

Permitted value Default Checked at

True/false

False

Startup

ssl.key.alias

Alias of private key / certificate chain of the server used to encrypt communications.

Permitted value Default Checked at

Alias of a keystore entry.

N/A

Startup

ssl.key.password

Encrypted password of private key entry in keystore. Can be encrypted using encrypt-password.sh.

Permitted value Default Checked at

Encrypted password

N/A

Startup

ssl.keystore

Local filesystem path of key store containing key entry.

Permitted value Default Checked at

Absolute path to key store.

N/A

Startup

ssl.keystore.password

Encrypted password of key store. Can be encrypted using encrypt-password.sh.

Permitted value Default Checked at

Encrypted password

N/A

Startup

ssl.truststore

Local filesystem path of trust store used to validate certificates sent by other Fusion Servers or IHC servers.

Permitted value Default Checked at

Absolute path to trust store.

N/A

Startup

ssl.truststore.password

Encrypted password of trust store. Can be encrypted using encrypt-password.sh.

Permitted value Default Checked at

Encrypted password

N/A

Startup

stacktrace.messaging.enabled

Allows you to turn off the ability to receive stack traces from REST API calls.

Permitted value Default Checked at

True/false

False

Startup

transfer.chunk.size

The size of the ChunkedStream.

Permitted value Default Checked at

1 – Integer.MAX_VALUE

32768

Startup

zone

The zone name for where the Fusion server is located. This is set during installation.

Permitted value Default Checked at

Any String (must not be altered)

N/A

Startup

Additional properties

The following properties are non-standard, and will not be present in a application.properties file by default.

agreed.proposal.store.rollback

Determines whether the Fusion server will attempt to rollback any uncommitted transactions on start up.

Permitted value Default Checked at

True/false

False

Startup

fusion.gsn.garbage.collector.thread.count

Coordinated requests (GSNs) are now removed from the shared storage using a thread pool. This controls the number of threads.

Permitted value Default Checked at

1 - MAX_INTEGER

10

Startup

fusion.gsn.garbage.collector.warn.threshold.time

If it takes longer than this time to garbage collect the coordinated requests (GSNs) stored in stable storage, then issue a warning and diagnostic event.

Permitted value Default Checked at

1 - MAX_LONG

30 (in seconds)

Startup

fusion.gsn.garbage.collector.block

Coordinated requests (GSNs) in stable storage are garbage collected by a thread pool, periodically requests will be submitted to garbage collect (GC) the GSNs. By default, the thread submitting requests does not wait for the GSN to be GC’d.
If this is set to true, then the thread that requests the GC will be blocked until the GC is done - this will cause back pressure on the system and reduce the throughput.

Permitted value Default Checked at

True/false

False

Startup

license.file

The path to the license file.

Permitted value Default Checked at

Absolute filesystem path

/etc/wandisco/fusion/server/license.key

Startup

netty.client.reuse.addr

Whether netty clients can reuse address and port for connections that are in the TIME_WAIT state.

Permitted value Default Checked at

True/false

False

Startup

read.timeout

Timeout of how long without response the IHC connection can be.

Permitted value Default Checked at

1 - MAX_LONG (in ms)

60000L

Startup

retry.sleep.time

The sleep time (milliseconds) in between retries of an agreed request.

Permitted value Default Checked at

1 - MAX_LONG (in ms)

1000L

Startup

ihc.connect.timeout

Timeout on how long Fusion should wait for IHC connection being established (networking timeout).

Permitted value Default Checked at

1 - MAX_LONG (in ms)

60000L

Startup

response.cache.size.limit

Allows you to set the size limit of the response cache which holds the processed request ids and corresponding responses.

Permitted value Default Checked at

1 - MAX_INTEGER

100000

Startup

response.cache.expiration

This specifies how long entries are stored in the response cache before they expire.

Permitted value Default Checked at

1 - MAX_INTEGER (in minutes)

60

Startup

7.1.2. IHC Server configuration

The Inter-Hadoop Communication Server is configured from a single file located at:

/etc/wandisco/fusion/ihc/server/{distro}/{version string}.ihc
Table 5. IHC Server properties
Property Description

http.server

The host and port for the web server, used when the ihc.http.policy is equal to HTTP_ONLY or BOTH_HTTP_HTTPS.

Permitted value Default Checked at

String:[1 - 65535]

0.0.0.0:9001

Startup

https.server

The host and port for the web server, used when the ihc.http.policy is equal to HTTPS_ONLY or BOTH_HTTP_HTTPS.

Permitted value Default Checked at

String:[1 - 65535]

0.0.0.0:8001

Startup

ihc.http.policy

Determines the HTTP policy supported by IHC Server.

Permitted value Default Checked at

HTTP_ONLY, HTTPS_ONLY, BOTH_HTTP_HTTPS

HTTP_ONLY

Startup

ihc.server.bind

The address the ihc server will bind to. If not specified, the default is "0.0.0.0:7000". In all cases, the port should be identical to the port used in the ihc.server address.

Permitted value Default Checked at

<String>:[1 - 65535]

0.0.0.0:7000

Startup

ihc.server

The hostname and port the IHC server will listen on.

Permitted value Default Checked at

<String>:[1 - 65535]

$FUSION_HOSTNAME:7000

Startup

ihc.ssl.enabled

Signifies that the IHC Server communications has SSL encryption enabled.

Permitted value Default Checked at

True/false

False

Startup

ihc.ssl.key.alias

Alias of private key / certificate chain of the IHC server used to encrypt communications.

Permitted value Default Checked at

Alias of a keystore entry

N/A

Startup

ihc.ssl.key.password

Encrypted password of private key entry in keystore. Can be encrypted using encrypt-password.sh.

Permitted value Default Checked at

Encrypted password

N/A

Startup

ihc.ssl.keystore

Local filesystem path of key store containing key entry.

Permitted value Default Checked at

Absolute path to key store.

N/A

Startup

ihc.ssl.keystore.password

Encrypted password of key store. Can be encrypted using encrypt-password.sh.

Permitted value Default Checked at

Encrypted password

N/A

Startup

ihc.ssl.keystore.type

Format of key store

Permitted value Default Checked at

JKS, PKCS12, etc.

JKS

Startup

ihc.ssl.truststore

Local filesystem path of trust store used to validate certificates sent by other IHC servers or Fusion Servers.

Permitted value Default Checked at

Absolute path to trust store

N/A

Startup

ihc.ssl.truststore.password

Encrypted password of trust store. Can be encrypted using encrypt-password.sh.

Permitted value Default Checked at

Encrypted password

N/A

Startup

ihc.ssl.truststore.type

Format of trust store.

Permitted value Default Checked at

JKS, PKCS12, etc.

JKS

Startup

Additional properties

The following properties are non-standard, and will not be present in a IHC properties file by default.

ihc.transfer.ssl.handshake.timeout

SSL Handshake timeout on transfer channel. Property added in 2.12.1.8.

Permitted value Default Checked at

Integer

60

Startup

ihc.transfer.write.limit

Write bandwidth limit on transfer channel. Property added in 2.12.1.8.

Permitted value Default Checked at

Integer (bytes/sec)

0 (Unlimited)

Startup

ihc.transfer.write.limit.check.interval

Check interval for bandwidth limit enforcement. Property added in 2.12.1.8.

Permitted value Default Checked at

Integer (seconds)

1

Startup

ihc.writer.threads

Number of threads servicing write handlers that perform reads from underlying storage and writes to network channel. Property added in 2.13.0.x

Permitted value Default Checked at

Integer

32

Startup

IHC Network configuration

The following is a description of how IHC servers are added to the replication system:

  1. The IHC servers are configured with the addresses of the WANdisco Fusion servers that inhabit the same zone.

  2. Periodically, the IHC servers ping the WANdisco Fusion servers using these stored addresses.

  3. The WANdisco Fusion servers will announce the IHC servers that have pinged them.

IHC servers in standard configuration should have the address of all WANdisco Fusion servers, since the core-site.xml property fusion.server lists them all. This is important because only the Writer node in each zone will confirm the existence of IHCs that have pinged it. Therefore the IHC Server has to talk to all Fusion servers in the zone in order to be flagged as available.

The same method is used in Hadoop to handle namenode and datanode connections. The datanode is configured with the namenode’s address and uses the address to contact the namenode and indicate its availability. If the namenode doesn’t hear from the datanode within a set period, the namenode assumes that the datanode is offline.

7.1.3. Fusion Client configuration

Usage Guide

There is a fixed relationship between the type of deployment and some of the Fusion Client parameters. The following table describes this relationship:

Table 6. Deployment type and Client Parameter combinations
Configuration fs.fusion.impl fs.AbstractFileSystem.fusion.impl fs.hdfs.impl

Use of fusion:/// with HCFS

com.wandisco.fs.client.FusionHcfs

com.wandisco.fs.client.FusionAbstractFs

Blank

Use of fusion:/// with HDFS

com.wandisco.fs.client.FusionHdfs

com.wandisco.fs.client.FusionAbstractFs

Blank

Use of hdfs:/// with HDFS

Blank

Blank

com.wandisco.fs.client.FusionHdfs

Use of fusion:/// and hdfs:/// with HDFS

com.wandisco.fs.client.FusionHdfs

com.wandisco.fs.client.FusionAbstractFs

com.wandisco.fs.client.FusionHdfs

Configuring properties for the Fusion Client

From 2.14, an additional file can be used to configure the Fusion Client. Using this file you can make updates to Fusion Client properties, without editing the core-site.xml which would require redistributing the core-site.xml and restarting Hadoop services.

The config.properties file can be placed in one of three locations. The client checks these in the following order, stopping at the first one found.

  • the replicated exchange directory i.e. hdfs://<fusion.replicated.dir.exchange>/<fusion.client.configuration.path>

  • the root of the Hadoop filesystem, i.e. hdfs:///<fusion.client.configuration.path>

  • on the local filesystem relative to where the process runs, i.e. /var/lib/hadoop-hdfs/<fusion.client.configuration.path>. If placed here, it will only affect local clients unless the file is replicated.

The default value for fusion.client.configuration.path is .fusion/client/config.properties.

For example, if fusion.client.configuration.path = my-configuration.properties, then the locations checked would be hdfs://<fusion.replicated.dir.exchange>/my-configuration.properties, hdfs://my-configuration.properties or /var/lib/hadoop-hdfs/my-configuration.properties.

If a file exists at any of the locations, it will be read as a properties file. Any of the Fusion Client properties that can be set in the core-site.xml can be set here. The expected keys and values are the same, and properties that can have multiple values should be comma-separated.

To change the system property fusion.client.configuration.path, set the value when running the client:

  • if you are using the hdfs client, set HADOOP_CLIENT_OPTS="-Dfusion.client.configuration.path=/the/path/to/my/config" and then run the client.

  • if you are using a pure java client, specify -Dfusion.client.configuration.path=/the/path/to/my/config when running the client.

7.1.4. Core-site configuration

Common location for HDP

/etc/hadoop/conf/core-site.xml

Common location for CDH

/etc/alternatives/hadoop-conf/core-site.xml

Common location for Cloud/LocalFS

Fusion Server = /etc/wandisco/fusion/server/core-site.xml
IHC Server = /etc/wandisco/fusion/ihc/server/{distro}/core-site.xml

The core-site.xml file contains many configurable properties that are used by the Fusion Client, Fusion Server, or both. These are described in this section, grouped by function.

The default for each property, where appropriate, is indicated in bold.

The Fusion Client properties can also be set in config.properties.

General settings
Table 7. Fusion properties in core-site.xml
Property Description

fs.AbstractFileSystem.fusion.impl

The Abstract FileSystem implementation to be used. Will not be present if "HDFS URI with HDFS" scheme is used. See the Fusion Client Usage Guide for details.

Permitted value Default Checked at

Java Class (string)

com.wandisco.fs.client.FusionAbstractFs

Startup

fs.fusion.impl

The FileSystem implementation to be used. Will not be present if "HDFS URI with HDFS" scheme is used. See the Fusion Client Usage Guide for details.

Permitted value Default Checked at

Java Class (string)

com.wandisco.fs.client.FusionHdfs

Startup

fs.fusion.push.threshold

The number of bytes the client will write before sending a push request to the Fusion server indicating bytes are available for transfer. If the threshold is 0, pushes are disabled. The default is the block size of the underlying filesystem.

Permitted value Default Checked at

0 - Long.MAX_VALUE

Block size of underlying FS

Startup

fs.fusion.underlyingFs

The address of the underlying filesystem. This is often the same as the fs.defaultFS property of the underlying hadoop. However, in cases like EMRFS, the fs.defaultFS points to a local HDFS built on the instance storage which is temporary, with persistent data being stored in S3. In this case, S3 storage is likely to be the fs.fusion.underlyingFs.

Permitted value Default Checked at

String:[1 – 65535] or URL

N/A

Startup

fs.hdfs.impl

The DistributedFileSystem implementation to be used. Will only be seen when certain schemes are in configured - "HDFS URI with HDFS" and "Fusion and HDFS URIs with HDFS". See the Fusion Client Usage Guide for further details.

Permitted value Default Checked at

Java Class (string)

com.wandisco.fs.client.FusionHdfs

Startup

fusion.dsmToken.dir

Customize .fusion dsm token directory. Without this property, the default is the replicated directory location. See section further below for more details.

Permitted value Default Checked at

Underlying filesystem path not under replication

N/A

Startup

fusion.replicated.dir.exchange

Location from which clients should try to read information about replicated directories, before contacting Fusion server. It’s necessary to configure the same in server’s application.properties, so that it generates the necessary data.

Permitted value Default Checked at

Underlying filesystem path not under replication

/wandisco/exchange_dir

Startup

fusion.server

The hostname and request port of the Fusion server. Comma-separated list of hostname:port for multiple Fusion servers in the Zone.

Permitted value Default Checked at

String:[1 – 65535]

N/A

Startup

Additional properties

The following properties are non-standard, and will not be present in a core-site.xml file by default.

fusion.client.retry.max.attempts

Max number of times to attempt to connect to a Fusion server before failing over (in the case of multiple Fusion servers).

Permitted value Default Checked at

Any integer

3

Startup

fs.fusion.underlyingFsClass

The name of the implementation class for the underlying file system specified with fs.fusion.underlyingFs.

Value

Fusion expects particular implementation classes to be associated with common URI schemes used by Hadoop clients when accessing the file system (e.g. s3://, file://, hdfs://, etc.)

If your cluster is configured to use alternative implementations classes for the scheme configured in fs.fusion.underlyingFs, you need to specify the name of the implementation for the underlying file system with this item. You also need to specify the implementation if using a URI scheme that is not one of those known to the defaults here.

Note - if you wish to write checksum (.crc) files to a local filesystem, this property should be change to org.apache.hadoop.fs.LocalFileSystem.

Default

There is a default per scheme:
abfs: org.apache.hadoop.fs.azurebfs.AzureBlobFileSystem
adl: org.apache.hadoop.fs.adl.AdlFileSystem
file: org.apache.hadoop.fs.RawLocalFileSystem
gs: com.google.cloud.hadoop.fs.gcs.GoogleHadoopFileSystem
hdfs: org.apache.hadoop.hdfs.DistributedFileSystem
mapr: com.mapr.fs.MapRFileSystem
s3: com.amazon.ws.emr.hadoop.fs.EmrFileSystem
s3a: org.apache.hadoop.fs.s3a.S3AFileSystem
s3n: com.amazon.ws.emr.hadoop.fs.EmrFileSystem
wasb: org.apache.hadoop.fs.azure.NativeAzureFileSystem
wasbs: org.apache.hadoop.fs.azure.NativeAzureFileSystem$Secure - if you get a ClassNoFoundExeption error see the workaround detailed here.

Checked at

Startup

fusion.enable.early.pulls

If true, the Fusion server will pull data when it receives an HFlush request from another zone.

Permitted value Default Checked at

True/false

False

When an HFlushRequest is received

fusion.client.connect.timeout.secs

The maximum period of time, in seconds, that the Fusion Client should spend attempting to establish a connection to each Fusion Server.

Permitted value Default Checked at

Any integer

60

Startup

fusion.client.thread.pool.size

The number of threads to use for the event loop group, per client unless the group is shared.

Permitted value Default Checked at

Any integer, optimised for powers of 2.

1

Startup

fusion.client.share.event.loop.group

A flag to indicate if unique instances of Fusion Client should share a common NettyEventLoopGroup.

Permitted value Default Checked at

True/false

False

Startup

fusion.client.request.ack.timeout.secs

The time before a Fusion Client request will timeout due to a lack of acknowledgement from the Fusion server.

Permitted value Default Checked at

Any integer

120

Startup

Details for fusion.dsmToken.dir property

This property is used to set a custom location for the .fusion DSM token directory. The property is not yet configurable in the Fusion UI. The following limitations apply:

  • The token directory can only be changed when no replicated directories exist. Otherwise the behavior is undefined.

  • Some files, such as the Consistency Check metadata are transferred via the standard IHC mechanism, which will require that the paths are identical across all zones. As a result, this configuration must be applied across all zones.

  • If the path used contains a special URL character, the characters must be encoded. For example, if the desired path is /my dsm&dir, it needs to be set as /my%20dsm%26dir.

Security Settings

The following properties are used if Kerberos and/or SSL are enabled on the underlying cluster. These properties must be defined in the /etc/hadoop/conf/core-site.xml file and if you are using unmanaged clusters will also need to be replicated to the IHC core-site.xml file. The default for each property, where appropriate, is indicated in bold.

Table 8. Fusion security properties in core-site.xml
Property Description

fusion.client.ssl.enabled

Enables SSL between the Fusion Server - Fusion Client. See Client to Fusion server encryption for details.

Permitted value Default Checked at

Usage

True/false

False

fusion.client.ssl.truststore

Local filesystem path of trust store used to validate certificates sent by Fusion Servers.

Permitted value Default Checked at

Absolute path to trust store

N/A

Startup

fusion.client.ssl.truststore.password

Encrypted password of trust store. Can be encrypted using encrypt-password.sh.

Permitted value Default Checked at

Encrypted password

N/A

Startup

fusion.client.ssl.truststore.type

Format of key store

Permitted value Default Checked at

JKS, PKCS12, etc.

JKS

Startup

fusion.handshakeToken.dir

Path to the Kerberos handshake token directory. The Fusion server(s) will attempt to write to this directory to verify that the user has the proper Kerberos credentials to write to the underlying file system. See Setting up handshake tokens for details of this property. If this property is not set, the default location for handshake tokens would be /user/$USERNAME.

Permitted value Default Checked at

Underlying Filesystem path

N/A

On processing a client request

fusion.http.authentication.enabled

Enables authentication on the REST API.

Permitted value Default Checked at

True/false

False

Startup

fusion.http.authentication.kerberos.keytab

The path to a keytab that contains the principal specified in fusion.http.authentication.kerberos.principal.

Permitted value Default Checked at

Absolute path to Kerberos keytab

N/A

Startup

fusion.http.authentication.kerberos.principal

The principal the fusion server will use to login with. The name of the principal must be "HTTP".

Permitted value Default Checked at

Kerberos principal in HTTP/_HOST@{KERBEROS_REALM} format

N/A

Startup

fusion.http.authentication.type

Type of authentication to use for http access.

Permitted value Default Checked at

simple or kerberos

simple

Startup

fusion.http.authorization.authorized.proxies

Users that are allowed to proxy on behalf of other users. HTTP calls would include a value for the header proxy.user.name. The proxied user’s permissions will then be checked against authorized readers and read-writers.

Permitted value Default Checked at

Comma-delimited list of users

HTTP

Startup

fusion.http.authorization.authorized.read.writers

Users that are allowed to make read OR write calls (any type of HTTP request).

Permitted value Default Checked at

Comma-delimited list of users

fusionUISystem

Startup

fusion.http.authorization.authorized.readers

Users that are allowed to make read calls ONLY (write calls are PATCH, POST, PUT, DELETE).

Permitted value Default Checked at

Comma-delimited list of users

fusionUISystem

Startup

fusion.http.authorization.enabled

Enables authorization on the REST API. Authentication must also be enabled.

Permitted value Default Checked at

True/false

False

Startup

fusion.keytab

The path to a keytab that contains the principal specified in fusion.principal.

Permitted value Default Checked at

Absolute path to Kerberos keytab

N/A

Startup

fusion.principal

The name of the fusion principal found in the fusion.keytab file. Used for Kerberos login purposes since a keytab can contain multiple principals.

Permitted value Default Checked at

{PRINCIPAL}@{KERBEROS_REALM}

N/A

Startup

Additional properties

The following properties are non-standard, and will not be present in a core-site.xml file by default.

fusion.system.user

If defined, Fusion impersonates to this user for performing various HDFS tasks that require system privileges (e.g. change of ownership, repairs, etc). It must be defined when fusion.principal is not a system user, otherwise the actions defining system privileges will fail. If not defined, the fusion.principal is used.

Permitted value Default Checked at

System Username

N/A

Startup

Consistency Check and Bypass Settings

The following properties determine how consistency checks and bypass run.

Table 9. Fusion Consistency checks and bypass properties in core-site.xml
Property Description

fusion.client.bypass.retry.interval.secs

The length of time for which the client will stay "bypassed" before trying Fusion again. This is will only occur if the Fusion server was still uncontactable.

Permitted value Default Checked at

Time in seconds

60

Every time a fusion client attempts to connect to a fusion server.

fusion.client.can.bypass

If the client cannot contact the Fusion server, this setting will allow a client request to bypass replication and proceed directly to the underlying filesystem after retrying.

Permitted value Default Checked at

True/false

False

Every time a client request is submitted

Additional properties

The following properties are non-standard, and will not be present in a core-site.xml file by default.

fusion.check.permissions

If true, then Fusion will include the permissions in the produced file statuses and they will then be compared in consistency check.
Default values
True for file://, hdfs://, maprfs://, wasb:// and wasbs://
False otherwise.

Permitted value Default Checked at

True/false

See note above

Startup

fusion.check.ownerAndGroup

If true, then Fusion will include the group and username in the produced file statuses and it will be then compared in consistency check.
For owner equivalence, this property needs to be true, as well as fusion.username.translations enabled. This will only be referenced if fusion.check.permissions=true.
Default values
True for file://, hdfs://, maprfs://, wasb:// and wasbs://
False otherwise.
See here if using ADL.

Permitted value Default Checked at

True/false

See note above

Startup

fusion.acls.supported

Whether Fusion will try to read or set ACLs. If false, they will not be compared in consistency check.

Permitted value Default Checked at

True/false

True for hdfs://, false otherwise.

Startup

fusion.append.supported

Determines whether the filesystem supports appends or not. Fusion already contains a list of filesystems that contain defaults based on whether appends are supported. For new filesystems, this configuration can be used.

Permitted value Default Checked at

True/false

default: "true" for new file systems

NA

LocalFileSystems

We’ve introduced FusionLocalFs for LocalFileSystems using WANdisco Fusion. This is necessary because there are a couple of places where the system expects to use a Local File System.

Configuration

fs.fusion.impl

fs.AbstractFileSystem.fusion.impl

fs.hdfs.impl

LocalFileSystems

com.wandisco.fs.client.FusionLocalFs

com.wandisco.fs.client.FusionLocalFs

com.wandisco.fs.client.FusionLocalFs

Therefore, for LocalFileSystems, users should set their fs.<parameter>.impl configuration to *com.wandisco.fs.client.FusionLocalFs*.

Usage
  • Set fs.file.impl to FusionLocalFs, (then any file:/// command will go through FusionLocalFs)

  • Set fs.fusion.impl to FusionLocalFs, (then any fusion:/// command will go through FusionLocalFs).

    Further more, a user can now set any scheme to any Fusion*Fs and when running a command with that scheme, it will go through that Fusion*Fs. e.g.,

  • Set fs.orange.impl to FusionLocalFs, (then any oranges:/// command will go through FusionLocalFs).

  • Set fs.lemon.impl to FusionHdfs, (then any lemon:/// command will go through FusionHdfs).

Azure Configuration

When using Fusion with Azure the following configurations can be added to /etc/hadoop/conf/core-site.xml or /etc/wandisco/fusion/server/core-site.xml if there is no Hadoop.

Table 10. Azure properties
Property Description

fusion.adl.ugi.check.path

The service principal user and group of the Filesystem instance is needed for ADL. This metadata is retrieved from the directory specified by this property. This directory requires write permissions by this client.

Permitted value Default Checked at

/a/b/c

The home directory i.e /user/hdfs

S3 Plugin Configuration

When using the Fusion S3 plugin the following configurations can be added to /etc/hadoop/conf/core-site.xml or /etc/wandisco/fusion/server/core-site.xml if there is no Hadoop.

Table 11. S3 Client plugin properties
Property Description

fs.fusion.s3.accelerateMode

Configures the client to use S3 accelerate endpoint for all requests.

Permitted value Default Checked at

True/false

False

Startup

fs.fusion.s3.bufferDir

Path to the directory where files are downloaded locally from IHC servers before uploading to s3. Multiple buffer directories are supported.

Permitted value Default Checked at

A path to a directory that is writable by the user running the fusion server.

System temp dir, usually /tmp.

Startup

fs.fusion.s3.bucketName

The name of the container that will store the file system.

Permitted value Default Checked at

The name of a s3 bucket.

None - must be present

Startup

fs.fusion.s3.cache.response.metadata

Sets whether or not to cache response metadata. Response metadata is typically used for troubleshooting issues with AWS support staff. While this feature is useful for debugging it adds overhead and so disabling it may be desired in high throughput applications. Property added in 2.12.3.

Permitted value Default Checked at

True/false

False

Startup

fs.fusion.s3.chunkedEncoding

Configures the client to enable chunked encoding automatically for PutObjectRequest and UploadPartRequest.

Permitted value Default Checked at

True/false

True

Startup

fs.fusion.s3.connectionTimeout

Time in milliseconds to wait when initially establishing a connection before timing out.

Permitted value Default Checked at

Time in milliseconds

10 seconds

Startup

fs.fusion.s3.endpoint

The s3 endpoint.

Permitted value Default Checked at